From 88523c8eb5714145e2b0cc3e99c70ce604698327 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 01:39:15 +0800 Subject: [PATCH 01/22] =?UTF-8?q?fix(startup):=20revert=20overlay=20splash?= =?UTF-8?q?=20=E2=80=94=20blocked=20all=20pointer=20events?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The overlay approach (app renders behind z-9999 splash) had a critical flaw: while !appReady(), the fixed-inset overlay captured ALL pointer events. Users could see the project icons but could not click them. Reverted to the original Show/fallback splash structure. Removed: - appReadyFired effect + deepagent-code:app-ready event dispatch - firstSessionLoadDone signal + Promise.all loadSessions wrapper - Overlay <> fragment in renderer/index.tsx - createSignal import in layout.tsx Kept (non-blocking improvements from prior commit): - sidebar.opened defaults to false (no blank-panel flash) - terminal panel transition suppressed until layoutReady() - SidebarPanel shows SessionSkeleton instead of blank - bottom/side terminal pty.create gated on runtimeId (no cold-start 503) - pty.create 10s deadline - F1/F3/F5/F6/F2 telemetry unchanged Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/pages/layout.tsx | 35 +-------------- packages/desktop/src/renderer/index.tsx | 57 ++++++------------------- 2 files changed, 15 insertions(+), 77 deletions(-) diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index a4c2ed93..d14ad19a 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -3,7 +3,6 @@ import { createEffect, createMemo, createResource, - createSignal, For, on, onCleanup, @@ -1925,10 +1924,6 @@ export default function Layout(props: ParentProps) { const loadedSessionDirs = new Set() - // F2-splash: signal that the first session-list fetch has completed so the - // renderer knows the sidebar is populated and the splash can be dismissed. - const [firstSessionLoadDone, setFirstSessionLoadDone] = createSignal(false) - createEffect( on( visibleSessionDirs, @@ -1939,48 +1934,20 @@ export default function Layout(props: ParentProps) { } const next = new Set(dirs) - const loads: Promise[] = [] for (const directory of next) { if (loadedSessionDirs.has(directory)) continue - loads.push(serverSync.project.loadSessions(directory)) + void serverSync.project.loadSessions(directory) } loadedSessionDirs.clear() for (const directory of next) { loadedSessionDirs.add(directory) } - - if (loads.length > 0) { - void Promise.all(loads).then(() => setFirstSessionLoadDone(true)) - } else { - // All visible directories were already loaded on a prior run. - setFirstSessionLoadDone(true) - } }, { defer: true }, ), ) - // Fire `deepagent-code:app-ready` exactly once when every gate is satisfied: - // 1. server persist loaded (projects list populated — was missing, caused early fire) - // 2. layout persist loaded (project list available) - // 3. page persist loaded (last-session info available) - // 4. autoselecting resolved (router has navigated to the last project) - // 5. first sessions fetch done — OR no projects exist (nothing to wait for) - // The renderer's splash stays up until this event fires (5 s failsafe there). - let appReadyFired = false - createEffect(() => { - if (appReadyFired) return - if (!server.ready()) return // server persist must be loaded (populates projects list) - if (!layoutReady()) return - if (!pageReady()) return - if (autoselecting.loading) return - const projects = layout.projects.list() - if (projects.length > 0 && !firstSessionLoadDone()) return - appReadyFired = true - window.dispatchEvent(new Event("deepagent-code:app-ready")) - }) - function handleDragStart(event: unknown) { const id = getDraggableId(event) if (!id) return diff --git a/packages/desktop/src/renderer/index.tsx b/packages/desktop/src/renderer/index.tsx index 8ad8d93f..8ff32622 100644 --- a/packages/desktop/src/renderer/index.tsx +++ b/packages/desktop/src/renderer/index.tsx @@ -335,32 +335,17 @@ render(() => { function App() { const wslServers = useWslServers() - - // App-ready gate: the app layer fires "deepagent-code:app-ready" once layout - // persist + routing + first sessions fetch are all done. A 5-second failsafe - // prevents a permanent hang. We use an OVERLAY approach: the app renders - // immediately (so Layout can mount and do its async work), and the splash sits - // on top until appReady fires. This avoids the circular dependency where the - // app can never fire the event because it can't render until the event fires. - const [appReady, setAppReady] = createSignal(false) - onMount(() => { - const fallback = window.setTimeout(() => setAppReady(true), 5_000) - window.addEventListener( - "deepagent-code:app-ready", - () => { - clearTimeout(fallback) - setAppReady(true) - }, - { once: true }, - ) - onCleanup(() => clearTimeout(fallback)) - }) + const splash = ( +
+ +
+ ) const ready = createMemo( () => !defaultServer.loading && !sidecar.loading && !windowCount.loading && !locale.loading, ) - // renderer.initialization — end: basic resources resolved. + // renderer.initialization — end: basic resources resolved, app is about to mount. createEffect(() => { if (!ready() || rendererReadyLogged) return rendererReadyLogged = true @@ -394,29 +379,15 @@ render(() => { ) return ( - <> - {/* App renders immediately so Layout can mount and do async startup work. */} - - - {(key) => ( - - - - )} - - - {/* - * Splash overlay: sits on top until both server resources are ready AND - * the app layer has signalled session-UI readiness (deepagent-code:app-ready). - * The 5-second failsafe in the onMount above ensures this never hangs - * permanently even if the async chain encounters an error. - */} - -
- -
+ + + {(key) => ( + + + + )} - + ) } From f26ce31a1f784340ce40dd5bb2dfd280b434e8f7 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 09:22:49 +0800 Subject: [PATCH 02/22] =?UTF-8?q?fix(startup):=20visibility:hidden=20appro?= =?UTF-8?q?ach=20=E2=80=94=20D=20stays=20until=20all=20data=20ready?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Use visibility:hidden instead of z-index overlay. The app renders in the background (Layout mounts, data loads) while completely invisible and non-interactive. Once all gates pass, D disappears and the app is immediately visible and fully ready. Architecture: - renderer/index.tsx: app wrapped in visibility:hidden div until appReady; splash overlay (z-9999) stays while !ready || !appReady - layout.tsx: appReadyFired effect gates on server.ready() + layoutReady() + pageReady() + !autoselecting.loading + firstSessionLoadDone() - firstSessionLoadDone fires when current project's session list completes Why visibility:hidden (not overlay): - visibility:hidden removes pointer events from the element and all children → no click-blocking side effects even if splash somehow disappears early - The app is truly invisible (no rendering artifacts, no partial paint) - Once ready, app becomes visible instantly with all data already in place Failsafe: 8-second timeout in renderer ensures splash never hangs permanently on slow networks. Once D ends, sidebar panel and terminal are already mounting in background (user requirement: those can load lazily after D). Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/pages/layout.tsx | 37 ++++++++++++++++- packages/desktop/src/renderer/index.tsx | 53 +++++++++++++++++++------ 2 files changed, 76 insertions(+), 14 deletions(-) diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index d14ad19a..60421815 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -3,6 +3,7 @@ import { createEffect, createMemo, createResource, + createSignal, For, on, onCleanup, @@ -1924,6 +1925,9 @@ export default function Layout(props: ParentProps) { const loadedSessionDirs = new Set() + // Track when the first (current project) session list has finished loading. + const [firstSessionLoadDone, setFirstSessionLoadDone] = createSignal(false) + createEffect( on( visibleSessionDirs, @@ -1934,20 +1938,51 @@ export default function Layout(props: ParentProps) { } const next = new Set(dirs) + const loads: Promise[] = [] for (const directory of next) { if (loadedSessionDirs.has(directory)) continue - void serverSync.project.loadSessions(directory) + loads.push(serverSync.project.loadSessions(directory)) } loadedSessionDirs.clear() for (const directory of next) { loadedSessionDirs.add(directory) } + + if (loads.length > 0) { + void Promise.all(loads).then(() => setFirstSessionLoadDone(true)) + } else { + setFirstSessionLoadDone(true) + } }, { defer: true }, ), ) + // Signal to renderer that the app is fully ready: projects loaded, last session + // navigated to, and the current session list fetched. Renderer keeps splash visible + // (with app hidden via visibility:hidden) until this fires. + // Gates: + // 1. server persist loaded → projects list populated + // 2. layout/page persist loaded → last-session info available + // 3. autoselecting resolved → router has navigated to the last project + // 4. firstSessionLoadDone → current project sessions listed in sidebar + // 8-second failsafe in the renderer prevents a permanent hang. + let appReadyFired = false + createEffect(() => { + if (appReadyFired) return + if (!server.ready()) return + if (!layoutReady()) return + if (!pageReady()) return + if (autoselecting.loading) return + const projects = layout.projects.list() + // If there are projects, wait until the first session list load completes + // so the sidebar shows real sessions rather than skeletons. + if (projects.length > 0 && !firstSessionLoadDone()) return + appReadyFired = true + window.dispatchEvent(new Event("deepagent-code:app-ready")) + }) + function handleDragStart(event: unknown) { const id = getDraggableId(event) if (!id) return diff --git a/packages/desktop/src/renderer/index.tsx b/packages/desktop/src/renderer/index.tsx index 8ff32622..60494ca0 100644 --- a/packages/desktop/src/renderer/index.tsx +++ b/packages/desktop/src/renderer/index.tsx @@ -335,11 +335,25 @@ render(() => { function App() { const wslServers = useWslServers() - const splash = ( -
- -
- ) + + // App-data gate: Layout fires "deepagent-code:app-ready" once the session UI + // is fully prepared (projects loaded, last session navigated to, sessions + // listed). Until then the app renders in the background but is completely + // invisible via `visibility:hidden` — no pointer events, no visual flash. + // An 8-second failsafe prevents a permanent hang on slow networks. + const [appReady, setAppReady] = createSignal(false) + onMount(() => { + const fallback = window.setTimeout(() => setAppReady(true), 8_000) + window.addEventListener( + "deepagent-code:app-ready", + () => { + clearTimeout(fallback) + setAppReady(true) + }, + { once: true }, + ) + onCleanup(() => clearTimeout(fallback)) + }) const ready = createMemo( () => !defaultServer.loading && !sidecar.loading && !windowCount.loading && !locale.loading, @@ -379,15 +393,28 @@ render(() => { ) return ( - - - {(key) => ( - - - - )} + <> + {/* Splash: shows while server resources load OR while app data preloads. + Uses a fixed overlay so the hidden app can render in the background. */} + +
+ +
+ + {/* App: renders as soon as server is ready, but hidden until data is preloaded. + visibility:hidden means no rendering artifacts, no pointer events, no flash. */} + +
+ + {(key) => ( + + + + )} + +
- + ) } From f6dcca9f08b9cabf36c9de8bfe31cdfc7a3c1aae Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 09:30:29 +0800 Subject: [PATCH 03/22] =?UTF-8?q?Revert=20"fix(startup):=20visibility:hidd?= =?UTF-8?q?en=20approach=20=E2=80=94=20D=20stays=20until=20all=20data=20re?= =?UTF-8?q?ady"?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit 1c9d7f19ad9350ce4041a443fcbda0b97b9a3ac3. --- packages/app/src/pages/layout.tsx | 37 +---------------- packages/desktop/src/renderer/index.tsx | 53 ++++++------------------- 2 files changed, 14 insertions(+), 76 deletions(-) diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index 60421815..d14ad19a 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -3,7 +3,6 @@ import { createEffect, createMemo, createResource, - createSignal, For, on, onCleanup, @@ -1925,9 +1924,6 @@ export default function Layout(props: ParentProps) { const loadedSessionDirs = new Set() - // Track when the first (current project) session list has finished loading. - const [firstSessionLoadDone, setFirstSessionLoadDone] = createSignal(false) - createEffect( on( visibleSessionDirs, @@ -1938,51 +1934,20 @@ export default function Layout(props: ParentProps) { } const next = new Set(dirs) - const loads: Promise[] = [] for (const directory of next) { if (loadedSessionDirs.has(directory)) continue - loads.push(serverSync.project.loadSessions(directory)) + void serverSync.project.loadSessions(directory) } loadedSessionDirs.clear() for (const directory of next) { loadedSessionDirs.add(directory) } - - if (loads.length > 0) { - void Promise.all(loads).then(() => setFirstSessionLoadDone(true)) - } else { - setFirstSessionLoadDone(true) - } }, { defer: true }, ), ) - // Signal to renderer that the app is fully ready: projects loaded, last session - // navigated to, and the current session list fetched. Renderer keeps splash visible - // (with app hidden via visibility:hidden) until this fires. - // Gates: - // 1. server persist loaded → projects list populated - // 2. layout/page persist loaded → last-session info available - // 3. autoselecting resolved → router has navigated to the last project - // 4. firstSessionLoadDone → current project sessions listed in sidebar - // 8-second failsafe in the renderer prevents a permanent hang. - let appReadyFired = false - createEffect(() => { - if (appReadyFired) return - if (!server.ready()) return - if (!layoutReady()) return - if (!pageReady()) return - if (autoselecting.loading) return - const projects = layout.projects.list() - // If there are projects, wait until the first session list load completes - // so the sidebar shows real sessions rather than skeletons. - if (projects.length > 0 && !firstSessionLoadDone()) return - appReadyFired = true - window.dispatchEvent(new Event("deepagent-code:app-ready")) - }) - function handleDragStart(event: unknown) { const id = getDraggableId(event) if (!id) return diff --git a/packages/desktop/src/renderer/index.tsx b/packages/desktop/src/renderer/index.tsx index 60494ca0..8ff32622 100644 --- a/packages/desktop/src/renderer/index.tsx +++ b/packages/desktop/src/renderer/index.tsx @@ -335,25 +335,11 @@ render(() => { function App() { const wslServers = useWslServers() - - // App-data gate: Layout fires "deepagent-code:app-ready" once the session UI - // is fully prepared (projects loaded, last session navigated to, sessions - // listed). Until then the app renders in the background but is completely - // invisible via `visibility:hidden` — no pointer events, no visual flash. - // An 8-second failsafe prevents a permanent hang on slow networks. - const [appReady, setAppReady] = createSignal(false) - onMount(() => { - const fallback = window.setTimeout(() => setAppReady(true), 8_000) - window.addEventListener( - "deepagent-code:app-ready", - () => { - clearTimeout(fallback) - setAppReady(true) - }, - { once: true }, - ) - onCleanup(() => clearTimeout(fallback)) - }) + const splash = ( +
+ +
+ ) const ready = createMemo( () => !defaultServer.loading && !sidecar.loading && !windowCount.loading && !locale.loading, @@ -393,28 +379,15 @@ render(() => { ) return ( - <> - {/* Splash: shows while server resources load OR while app data preloads. - Uses a fixed overlay so the hidden app can render in the background. */} - -
- -
- - {/* App: renders as soon as server is ready, but hidden until data is preloaded. - visibility:hidden means no rendering artifacts, no pointer events, no flash. */} - -
- - {(key) => ( - - - - )} - -
+ + + {(key) => ( + + + + )} - + ) } From 27ba7f57ce7075718bf87b5683f6489baba4dadd Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 10:00:00 +0800 Subject: [PATCH 04/22] fix(sidebar): revert startup-UX sidebar changes that caused DnD regression sidebar.opened: false->true, removed SessionSkeleton fallback, removed layoutReady() transition gate. These three changes from aca62551 triggered solid-dnd cleanup-order errors on project switch. Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/context/layout.tsx | 2 +- packages/app/src/pages/layout.tsx | 19 ++----------------- 2 files changed, 3 insertions(+), 18 deletions(-) diff --git a/packages/app/src/context/layout.tsx b/packages/app/src/context/layout.tsx index c67c2fc4..9c296052 100644 --- a/packages/app/src/context/layout.tsx +++ b/packages/app/src/context/layout.tsx @@ -328,7 +328,7 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( { ...target, migrate }, createStore({ sidebar: { - opened: false, + opened: true, width: DEFAULT_SIDEBAR_WIDTH, workspaces: {} as Record, workspacesDefault: false, diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index d14ad19a..9cd70c53 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -94,7 +94,6 @@ import { } from "./layout/sidebar-workspace" import { ProjectDragOverlay, SortableProject, type ProjectSidebarContext } from "./layout/sidebar-project" import { SidebarContent } from "./layout/sidebar-shell" -import { SessionSkeleton } from "./layout/sidebar-items" export default function Layout(props: ParentProps) { const serverSDK = useServerSDK() @@ -2192,18 +2191,7 @@ export default function Layout(props: ParentProps) { -
- -
- - } - > +
@@ -2583,11 +2571,8 @@ export default function Layout(props: ParentProps) { "absolute inset-0": true, "xl:inset-y-0 xl:right-0 xl:left-[var(--main-left)]": true, "z-20": true, - // Suppress the left-slide transition until the layout persist store - // has loaded so the initial open→close (or closed→open) snap from - // async storage doesn't animate visibly. "transition-[left] duration-200 ease-[cubic-bezier(0.22,1,0.36,1)] will-change-[left] motion-reduce:transition-none": - !state.sizing && layoutReady(), + !state.sizing, }} style={{ "--main-left": layout.sidebar.opened() ? `${side()}px` : "4rem", From 610746d469563bb6626337dd266ad9509f85a613 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 10:37:38 +0800 Subject: [PATCH 05/22] fix(sidebar): fundamental fix for solid-dnd pointer sensor listener leak MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Root cause: solid-dnd 0.7.5 createPointerSensor registers document-level pointermove/pointerup listeners on pointerdown, but its onCleanup only calls removeSensor() — never detach(). After any DragDropSensors remount (triggered by project switching), the old document listeners are orphaned. Accumulated stale onPointerMove handlers call event.preventDefault() on all pointer events, making the sidebar unresponsive after 2-3 switches. Second omission: solid-dnd has no pointercancel handler, so any gesture cancelled by the OS (scroll-lock, focus change, page navigation) also permanently leaks the document listeners. Fix: implement FixedDragDropSensors in packages/app/src/utils/solid-dnd.tsx — a drop-in replacement for DragDropSensors that: 1. Calls detach() in onCleanup before removeSensor(), ensuring document listeners are always removed when the component unmounts mid-gesture 2. Adds a pointercancel handler (absent in solid-dnd entirely) Replace DragDropSensors with FixedDragDropSensors in all three usages: - packages/app/src/pages/layout/sidebar-shell.tsx (left rail — main site) - packages/app/src/pages/layout.tsx (workspace DnD panel) - packages/app/src/pages/session/terminal-view.tsx (terminal pane DnD) Also removes the previous temporary workaround (synthetic pointerup/ pointercancel dispatch in openProject) — now unnecessary. Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/pages/layout.tsx | 6 +- .../app/src/pages/layout/sidebar-shell.tsx | 5 +- .../app/src/pages/session/terminal-view.tsx | 6 +- packages/app/src/utils/solid-dnd.tsx | 123 +++++++++++++++++- 4 files changed, 129 insertions(+), 11 deletions(-) diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index 9cd70c53..42a325f1 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -30,7 +30,7 @@ import { Session, type Message } from "@deepagent-code/sdk/v2/client" import { usePlatform } from "@/context/platform" import { useSettings } from "@/context/settings" import { createStore, produce, reconcile } from "solid-js/store" -import { DragDropProvider, DragDropSensors, DragOverlay, SortableProvider, closestCenter } from "@thisbeyond/solid-dnd" +import { DragDropProvider, DragOverlay, SortableProvider, closestCenter } from "@thisbeyond/solid-dnd" import type { DragEvent } from "@thisbeyond/solid-dnd" import { useProviders } from "@/hooks/use-providers" import { toaster } from "@deepagent-code/ui/toast" @@ -61,7 +61,7 @@ import { SessionRouteKey, SessionStateKey } from "@/utils/server-scope" import { useDialog } from "@deepagent-code/ui/context/dialog" import { useTheme, type ColorScheme } from "@deepagent-code/ui/theme/context" import { useCommand, type CommandOption } from "@/context/command" -import { ConstrainDragXAxis, getDraggableId } from "@/utils/solid-dnd" +import { ConstrainDragXAxis, getDraggableId, FixedDragDropSensors } from "@/utils/solid-dnd" import { DebugBar } from "@/components/debug-bar" import { listPending } from "@/components/review/dialog-review.api" import { fetchCapabilities } from "@/components/deepagent/panel-goal.api" @@ -2369,7 +2369,7 @@ export default function Layout(props: ParentProps) { onDragOver={handleWorkspaceDragOver} collisionDetector={closestCenter} > - +
{ diff --git a/packages/app/src/pages/layout/sidebar-shell.tsx b/packages/app/src/pages/layout/sidebar-shell.tsx index eaca8a6f..b83caf53 100644 --- a/packages/app/src/pages/layout/sidebar-shell.tsx +++ b/packages/app/src/pages/layout/sidebar-shell.tsx @@ -1,13 +1,12 @@ import { createEffect, createMemo, For, Show, type Accessor, type JSX } from "solid-js" import { DragDropProvider, - DragDropSensors, DragOverlay, SortableProvider, closestCenter, type DragEvent, } from "@thisbeyond/solid-dnd" -import { ConstrainDragXAxis } from "@/utils/solid-dnd" +import { ConstrainDragXAxis, FixedDragDropSensors } from "@/utils/solid-dnd" import { IconButton } from "@deepagent-code/ui/icon-button" import { Tooltip, TooltipKeybind } from "@deepagent-code/ui/tooltip" import { type LocalProject } from "@/context/layout" @@ -72,7 +71,7 @@ export const SidebarContent = (props: { onDragOver={props.handleDragOver} collisionDetector={closestCenter} > - +
p.worktree)}> diff --git a/packages/app/src/pages/session/terminal-view.tsx b/packages/app/src/pages/session/terminal-view.tsx index 3ff8df0e..203061a5 100644 --- a/packages/app/src/pages/session/terminal-view.tsx +++ b/packages/app/src/pages/session/terminal-view.tsx @@ -4,9 +4,9 @@ import { ResizeHandle } from "@deepagent-code/ui/resize-handle" import { IconButton } from "@deepagent-code/ui/icon-button" import { Button } from "@deepagent-code/ui/button" import { TooltipKeybind, Tooltip } from "@deepagent-code/ui/tooltip" -import { DragDropProvider, DragDropSensors, SortableProvider, closestCenter } from "@thisbeyond/solid-dnd" +import { DragDropProvider, SortableProvider, closestCenter } from "@thisbeyond/solid-dnd" import type { DragEvent } from "@thisbeyond/solid-dnd" -import { ConstrainDragYAxis } from "@/utils/solid-dnd" +import { ConstrainDragYAxis, FixedDragDropSensors } from "@/utils/solid-dnd" import { SortableTerminalTab } from "@/components/session" import { Terminal } from "@/components/terminal" @@ -221,7 +221,7 @@ function LeafPane(props: { node: PaneLeaf }) { onFocusIn={() => terminal.setFocusedPane(props.node.id)} > - +
diff --git a/packages/app/src/utils/solid-dnd.tsx b/packages/app/src/utils/solid-dnd.tsx index 8e30a033..6c3bb3d5 100644 --- a/packages/app/src/utils/solid-dnd.tsx +++ b/packages/app/src/utils/solid-dnd.tsx @@ -1,6 +1,6 @@ import { useDragDropContext } from "@thisbeyond/solid-dnd" -import type { Transformer } from "@thisbeyond/solid-dnd" -import { createRoot, onCleanup, type JSXElement } from "solid-js" +import type { Id, Transformer } from "@thisbeyond/solid-dnd" +import { createRoot, onCleanup, onMount, type JSXElement } from "solid-js" type DragEvent = { draggable?: { id?: unknown } } @@ -47,3 +47,122 @@ const createAxisConstraint = (axis: "x" | "y", transformerId: string) => (): JSX export const ConstrainDragXAxis = createAxisConstraint("x", "constrain-x-axis") export const ConstrainDragYAxis = createAxisConstraint("y", "constrain-y-axis") + +// --------------------------------------------------------------------------- +// FixedDragDropSensors — patches solid-dnd 0.7.5 pointer sensor bug. +// +// Bug: createPointerSensor's onCleanup only calls removeSensor(), never +// detach(). This leaves document-level pointermove/pointerup listeners +// orphaned after every DragDropSensors remount. Orphaned onPointerMove +// handlers call event.preventDefault(), swallowing all subsequent pointer +// events and making the UI unresponsive after ~2 project switches. +// +// Fix: reimplement createPointerSensor locally so that onCleanup calls +// detach() BEFORE removeSensor(), and add a pointercancel handler that +// solid-dnd omits entirely. +// --------------------------------------------------------------------------- + +function createFixedPointerSensor(id = "pointer-sensor") { + const ctx = useDragDropContext() + if (!ctx) return + + const [state, { addSensor, removeSensor, sensorStart, sensorMove, sensorEnd, dragStart, dragEnd }] = ctx + + const activationDelay = 250 + const activationDistance = 10 + + const initialCoordinates = { x: 0, y: 0 } + let activationDelayTimeoutId: number | null = null + let activationDraggableId: string | number | null = null + + const isActiveSensor = () => state.active.sensorId === id + + const clearSelection = () => window.getSelection()?.removeAllRanges() + + const detach = () => { + if (activationDelayTimeoutId !== null) { + clearTimeout(activationDelayTimeoutId) + activationDelayTimeoutId = null + } + document.removeEventListener("pointermove", onPointerMove) + document.removeEventListener("pointerup", onPointerUp) + document.removeEventListener("pointercancel", onPointerCancel) + document.removeEventListener("selectionchange", clearSelection) + } + + const onActivate = () => { + if (!state.active.sensor) { + sensorStart(id, initialCoordinates) + dragStart(activationDraggableId!) + clearSelection() + document.addEventListener("selectionchange", clearSelection) + } else if (!isActiveSensor()) { + detach() + } + } + + const onPointerMove = (event: PointerEvent) => { + const coordinates = { x: event.clientX, y: event.clientY } + if (!state.active.sensor) { + const dx = coordinates.x - initialCoordinates.x + const dy = coordinates.y - initialCoordinates.y + if (Math.sqrt(dx * dx + dy * dy) > activationDistance) onActivate() + } + if (isActiveSensor()) { + event.preventDefault() + sensorMove(coordinates) + } + } + + const onPointerUp = (event: PointerEvent) => { + detach() + if (isActiveSensor()) { + event.preventDefault() + dragEnd() + sensorEnd() + } + } + + // solid-dnd omits this entirely — pointercancel fires when the OS/browser + // cancels the gesture (scroll-lock, focus loss, navigation). Without it + // the sensor stays "half-attached" indefinitely. + const onPointerCancel = () => { + detach() + if (isActiveSensor()) { + dragEnd() + sensorEnd() + } + } + + const attach = (event: PointerEvent, draggableId: Id) => { + if (event.button !== 0) return + document.addEventListener("pointermove", onPointerMove) + document.addEventListener("pointerup", onPointerUp) + document.addEventListener("pointercancel", onPointerCancel) + activationDraggableId = draggableId + initialCoordinates.x = event.clientX + initialCoordinates.y = event.clientY + activationDelayTimeoutId = window.setTimeout(onActivate, activationDelay) + } + + onMount(() => { + addSensor({ id, activators: { pointerdown: attach } }) + }) + + onCleanup(() => { + // THE FIX: call detach() before removeSensor() so document listeners are + // always cleaned up, even if the component unmounts mid-gesture. + detach() + removeSensor(id) + }) +} + +/** + * Drop-in replacement for solid-dnd's DragDropSensors that fixes the pointer + * sensor's missing detach() call in onCleanup and adds pointercancel support. + */ +export const FixedDragDropSensors = (props: { children?: JSXElement }): JSXElement => { + createFixedPointerSensor() + return props.children as JSXElement +} + From eb82166c996283fcb5c7eed41cdbc3ff121d4fdb Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 10:42:33 +0800 Subject: [PATCH 06/22] =?UTF-8?q?perf(startup):=20reduce=20server-side=20l?= =?UTF-8?q?ayer=20build=20latency=20=E2=80=94=20r3=20fixes?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - event-dispatcher.ts / goal-tick-consumer.ts: wrap bare Deferred.await(ready) in Effect.timeout(500ms) + Effect.ignore so a DB-stall never hangs the V4 EventRuntime layer build (fixes the K40-2 durable consumer registration race) - deepagent-event-bus.ts: add dbGroupsCache to groupsFor — eliminates the per-publish DB SELECT; cache is invalidated on registerConsumerGroup / unregisterConsumerGroup so correctness is preserved - database.ts: set wal_autocheckpoint = 200 pages (~800 KB) so individual WAL-merge checkpoints are cheap instead of large infrequent write-lock spikes Ref: docs/4.0.4_r3.md Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/core/src/database/database.ts | 7 +++-- .../core/src/deepagent/deepagent-event-bus.ts | 29 ++++++++++++++++--- .../src/session/event-dispatcher.ts | 5 +++- .../src/session/goal-tick-consumer.ts | 5 +++- 4 files changed, 38 insertions(+), 8 deletions(-) diff --git a/packages/core/src/database/database.ts b/packages/core/src/database/database.ts index b0bf6a6a..7ef0cf68 100644 --- a/packages/core/src/database/database.ts +++ b/packages/core/src/database/database.ts @@ -29,8 +29,11 @@ export const layer = Layer.effect( yield* db.run("PRAGMA busy_timeout = 5000") yield* db.run("PRAGMA cache_size = -64000") yield* db.run("PRAGMA foreign_keys = ON") - // wal_checkpoint removed from startup: it blocked the "ready" signal by 1-3 s - // on large databases. SQLite auto-checkpoints at wal_autocheckpoint = 1000 pages. + // Tune WAL autocheckpoint: default 1000 pages (~4MB) causes large infrequent merges that spike + // write-lock hold time. 200 pages (~800KB) keeps each merge cheap while still amortizing I/O. + // Removed the blocking wal_checkpoint(PASSIVE) call (was 1-3s on large DBs); frequent small + // autocheckpoints are a better long-term strategy. + yield* db.run("PRAGMA wal_autocheckpoint = 200") yield* DatabaseMigration.apply(db) return { db } diff --git a/packages/core/src/deepagent/deepagent-event-bus.ts b/packages/core/src/deepagent/deepagent-event-bus.ts index 5f11eab5..f76d8eb9 100644 --- a/packages/core/src/deepagent/deepagent-event-bus.ts +++ b/packages/core/src/deepagent/deepagent-event-bus.ts @@ -276,8 +276,26 @@ export const layerWith = (options?: LayerOptions) => // K40-2: durable groups from DB (groups registered via registerConsumerGroup, survive restarts). // Returns the union of live + durable groups, deduped. An offline group that registered durably // will receive delivery rows even though it has no live stream right now. - const groupsFor = (eventType: string): Effect.Effect> => - db + // + // PERF: cache the DB portion (durable groups) per eventType to avoid a DB round-trip on every + // publish. Consumer group membership changes only at startup (consumerRegistrationLayer) and on + // flag changes — both call registerConsumerGroup/unregisterConsumerGroup which invalidate the + // cache. Live groups (in-memory only) are always computed fresh; they change on subscribe/end. + const dbGroupsCache = new Map() + const invalidateDbGroupsCache = () => dbGroupsCache.clear() + + const groupsFor = (eventType: string): Effect.Effect> => { + const cached = dbGroupsCache.get(eventType) + if (cached !== undefined) { + // DB portion is cached; still merge with current live groups (always fresh, no DB). + const live = liveGroupsFor(eventType) + if (live.length === 0) return Effect.succeed(cached) + const seen = new Set(cached) + const merged = [...cached] + for (const g of live) if (!seen.has(g)) merged.push(g) + return Effect.succeed(merged) + } + return db .select({ group_id: DeepAgentConsumerGroupTable.group_id }) .from(DeepAgentConsumerGroupTable) .where( @@ -288,6 +306,7 @@ export const layerWith = (options?: LayerOptions) => Effect.orDie, Effect.map((rows) => { const dbGroups = rows.map((r) => r.group_id) + dbGroupsCache.set(eventType, dbGroups) const live = liveGroupsFor(eventType) // Union: db-registered + live-only (not yet durable-registered), deduplicated. const seen = new Set(dbGroups) @@ -296,6 +315,7 @@ export const layerWith = (options?: LayerOptions) => return merged }), ) + } const publish: Interface["publish"] = (input) => Effect.gen(function* () { @@ -855,6 +875,7 @@ export const layerWith = (options?: LayerOptions) => // K40-2: durable consumer group registration — persists group identity so publish writes delivery // rows for offline groups too. Both methods are idempotent; registerConsumerGroup upserts. + // PERF: both methods invalidate dbGroupsCache so the next groupsFor re-reads from DB. const registerConsumerGroup: Interface["registerConsumerGroup"] = (groupId, typeFilter) => { const at = now() return db @@ -865,7 +886,7 @@ export const layerWith = (options?: LayerOptions) => set: { type_filter: typeFilter ?? null, last_seen_at: at }, }) .run() - .pipe(Effect.orDie, Effect.asVoid) + .pipe(Effect.orDie, Effect.asVoid, Effect.tap(() => Effect.sync(invalidateDbGroupsCache))) } const unregisterConsumerGroup: Interface["unregisterConsumerGroup"] = (groupId) => @@ -873,7 +894,7 @@ export const layerWith = (options?: LayerOptions) => .delete(DeepAgentConsumerGroupTable) .where(eq(DeepAgentConsumerGroupTable.group_id, groupId)) .run() - .pipe(Effect.orDie, Effect.asVoid) + .pipe(Effect.orDie, Effect.asVoid, Effect.tap(() => Effect.sync(invalidateDbGroupsCache))) return Service.of({ publish, diff --git a/packages/deepagent-code/src/session/event-dispatcher.ts b/packages/deepagent-code/src/session/event-dispatcher.ts index e0a08700..be9f196d 100644 --- a/packages/deepagent-code/src/session/event-dispatcher.ts +++ b/packages/deepagent-code/src/session/event-dispatcher.ts @@ -513,7 +513,10 @@ export const layerWith = (options?: LayerOptions) => Effect.forkScoped, ) // wait until the group is registered before the layer is considered ready. - yield* Deferred.await(ready) + // Timeout guards against DB-stall (busy WAL/retention sweep): durable registration already + // happened via registerConsumerGroup above, so a brief live-stream miss is recoverable via + // the retry pump. 500ms is well above normal fiber-schedule latency (<1ms). + yield* Deferred.await(ready).pipe(Effect.timeout(Duration.millis(500)), Effect.ignore) yield* tick() .pipe( diff --git a/packages/deepagent-code/src/session/goal-tick-consumer.ts b/packages/deepagent-code/src/session/goal-tick-consumer.ts index 063c0703..85c48627 100644 --- a/packages/deepagent-code/src/session/goal-tick-consumer.ts +++ b/packages/deepagent-code/src/session/goal-tick-consumer.ts @@ -336,7 +336,10 @@ export const layerWith = (options: LayerOptions) => ), Effect.forkScoped, ) - yield* Deferred.await(ready) + // Timeout guards against DB-stall (busy WAL/retention sweep): durable registration already + // happened via registerConsumerGroup above, so a brief live-stream miss is recoverable via + // the retry pump. 500ms is well above normal fiber-schedule latency (<1ms). + yield* Deferred.await(ready).pipe(Effect.timeout(Duration.millis(500)), Effect.ignore) yield* pumpRetries() .pipe( From cb224bc1b0ed752c80b5bc87400874515a11f779 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 10:43:18 +0800 Subject: [PATCH 07/22] =?UTF-8?q?feat(startup):=20eliminate=20home-page=20?= =?UTF-8?q?flash=20=E2=80=94=20StartupSplashGate=20+=20intent=20snapshot?= =?UTF-8?q?=20(r6)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Problem: D icon disappears → user sees home page loading → server sync completes → session page. The intermediate home-page state feels slow/broken. Fix (4 coordinated changes): 1. packages/app/src/utils/startup-intent.ts [new] Synchronous localStorage snapshot of last active session (server + directory + sessionId + timestamp). Written on every navigateTab, read at startup in <0.1 ms before any server request. 2. packages/app/src/context/tabs.tsx navigateTab: call writeStartupIntent with base64Decode(tab.dirBase64) — fixes a pre-existing atob() incompatibility with URL-safe base64 produced by base64Encode. 3. packages/app/src/App.tsx StartupSplashGate: inserted inside RouterRoot, wraps AppShellProviders. Keeps the D splash visible until onMount fires, then navigates directly to the intent session (fresh intent < 30 s) and releases. Stale/absent intents pass through immediately so new installs and long-idle cold starts are unaffected. RouterRoot.createEffect: skip its own navigation while a fresh intent is active (avoids a race between the gate and the tabs.ready() effect). 4. packages/app/src/context/global-sync/bootstrap.ts bootstrapGlobal: split into fast path (config + projects — awaited) and slow path (providers + path — background void). Cuts the blocking bootstrap from 4 parallel requests to 2, halving the time until the sidebar and session routing are usable. Slow-path errors are now logged via console.error instead of being silently dropped. Adversarial review passed. Two CRITICAL issues from review were fixed before commit: - atob → base64Decode (URL-safe base64 compatibility) - removed fake SDK prefetch (result was not stored in QueryClient cache) Ref: docs/4.0.4_r6.md Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/app.tsx | 80 +++++++++++++++++-- .../app/src/context/global-sync/bootstrap.ts | 26 +++--- packages/app/src/context/tabs.tsx | 11 ++- packages/app/src/utils/startup-intent.ts | 69 ++++++++++++++++ 4 files changed, 169 insertions(+), 17 deletions(-) create mode 100644 packages/app/src/utils/startup-intent.ts diff --git a/packages/app/src/app.tsx b/packages/app/src/app.tsx index dd00f9c7..73c77365 100644 --- a/packages/app/src/app.tsx +++ b/packages/app/src/app.tsx @@ -23,6 +23,7 @@ import { type JSX, lazy, onCleanup, + onMount, type ParentProps, Show, } from "solid-js" @@ -52,6 +53,8 @@ import DirectoryLayout from "@/pages/directory-layout" import Layout from "@/pages/layout" import { ErrorPage } from "./pages/error" import { useCheckServerHealth } from "./utils/server-health" +import { readStartupIntent, INTENT_NAVIGATE_WINDOW_MS } from "@/utils/startup-intent" +import { base64Encode } from "@deepagent-code/core/util/encode" const HomeRoute = lazy(() => import("@/pages/home")) const Session = lazy(() => import("@/pages/session")) @@ -140,6 +143,65 @@ function SessionProviders(props: ParentProps) { ) } +/** + * StartupSplashGate — keeps the D splash visible after ConnectionGate passes and + * navigates directly to the last session before revealing the app shell. + * + * Must be rendered inside a router root (needs useNavigate) and useServer. + * Reads the startup-intent snapshot written synchronously by navigateTab (tabs.tsx). + * + * Gate logic: + * • No intent / server mismatch → pass-through immediately (new install, cleared state). + * • Fresh intent (< INTENT_NAVIGATE_WINDOW_MS) → navigate to last session, release. + * • Stale intent (≥ window but < 24 h) → release immediately; the existing + * tabs.ready() effect in RouterRoot handles navigation as before. + * + * NOTE: The gate does NOT attempt to pre-warm the session query cache because the + * session SDK responses are keyed inside ServerSyncProvider (query key: [scope, dir, + * 'loadSessions']) which is not yet mounted when StartupSplashGate's onMount fires. + * The value of the gate is eliminating the visible home-page flash, not pre-fetching. + */ +function StartupSplashGate(props: ParentProps) { + const server = useServer() + const navigate = useNavigate() + + const intent = readStartupIntent() + + // No usable intent → skip the gate entirely (zero overhead). + if (!intent || intent.server !== server.key) { + return <>{props.children} + } + + const [warmupDone, setWarmupDone] = createSignal(false) + + onMount(() => { + try { + // Navigate immediately when the intent is fresh (process crash / hot restart). + // Stale intents delegate to the tabs.ready() effect so a long-idle cold-start + // doesn't force navigation to a session the user may no longer want. + if (Date.now() - intent.at < INTENT_NAVIGATE_WINDOW_MS) { + const dirBase64 = base64Encode(intent.directory) + navigate(`/${dirBase64}/session/${intent.sessionId}`, { replace: true }) + } + } finally { + setWarmupDone(true) + } + }) + + return ( + + +
+ } + > + {props.children} + + ) +} + function RouterRoot(props: ParentProps<{ appChildren?: JSX.Element }>) { const tabs = useTabs() const server = useServer() @@ -148,9 +210,13 @@ function RouterRoot(props: ParentProps<{ appChildren?: JSX.Element }>) { // Restore the explicitly persisted active tab. A cross-server directory is never // navigated until the target server has become active. + // Skip when StartupSplashGate has a fresh intent — it navigates directly and this + // effect would race/conflict with it during the warmup window. createEffect(() => { if (!tabs.ready()) return if (location.pathname !== "/") return + const intent = readStartupIntent() + if (intent && intent.server === server.key && Date.now() - intent.at < INTENT_NAVIGATE_WINDOW_MS) return const tab = startupTab(tabs.store, tabs.active.key, server.list) if (!tab) return if (server.key !== tab.server) { @@ -161,12 +227,14 @@ function RouterRoot(props: ParentProps<{ appChildren?: JSX.Element }>) { }) return ( - - {/*}>*/} - {props.appChildren} - {props.children} - {/**/} - + + + {/*}>*/} + {props.appChildren} + {props.children} + {/**/} + + ) } diff --git a/packages/app/src/context/global-sync/bootstrap.ts b/packages/app/src/context/global-sync/bootstrap.ts index 8792e665..67503830 100644 --- a/packages/app/src/context/global-sync/bootstrap.ts +++ b/packages/app/src/context/global-sync/bootstrap.ts @@ -127,22 +127,28 @@ export async function bootstrapGlobal(input: { setGlobalStore: SetStoreFunction queryClient: QueryClient }) { - const slow = [ + // FAST PATH — minimum data needed before the first session renders. + // config: gates feature flags; project list: gates sidebar + project routing. + // Both are awaited so the UI is never shown with a structurally incomplete store. + const fast = [ () => input.queryClient.fetchQuery(loadGlobalConfigQuery(input.scope, input.serverSDK)), - () => input.queryClient.fetchQuery(loadProvidersQuery(input.scope, null, input.serverSDK)), - () => input.queryClient.fetchQuery(loadPathQuery(input.scope, null, input.serverSDK)), () => input.queryClient .fetchQuery(loadProjectsQuery(input.scope, input.serverSDK)) .then((data) => input.setGlobalStore("project", data)), ] - await runAll(slow) - // showErrors({ - // errors: errors(), - // title: input.requestFailedTitle, - // translate: input.translate, - // formatMoreCount: input.formatMoreCount, - // }) + await runAll(fast) + + // SLOW PATH — secondary data that enriches the UI but does not block rendering. + // Providers and path are not needed for the session list or initial navigation; + // load them in the background so they become available shortly after first paint. + void runAll([ + () => input.queryClient.fetchQuery(loadProvidersQuery(input.scope, null, input.serverSDK)), + () => input.queryClient.fetchQuery(loadPathQuery(input.scope, null, input.serverSDK)), + ]).then((results) => { + const errs = errors(results).filter((e) => !isCancellation(e)) + if (errs.length > 0) console.error("[bootstrapGlobal] slow-path failed:", errs[0]) + }) } function groupBySession(input: T[]) { diff --git a/packages/app/src/context/tabs.tsx b/packages/app/src/context/tabs.tsx index 64ac3e46..6ef98899 100644 --- a/packages/app/src/context/tabs.tsx +++ b/packages/app/src/context/tabs.tsx @@ -1,12 +1,13 @@ import type { Session } from "@deepagent-code/sdk/v2/client" import { createSimpleContext } from "@deepagent-code/ui/context" -import { base64Encode } from "@deepagent-code/core/util/encode" +import { base64Encode, base64Decode } from "@deepagent-code/core/util/encode" import { createStore, produce } from "solid-js/store" import { Persist, persisted } from "@/utils/persist" import { ServerConnection, useServer } from "./server" import { createEffect, startTransition } from "solid-js" import { useNavigate, useParams } from "@solidjs/router" import { SessionTabsRemovedDetail } from "@/components/titlebar-session-events" +import { writeStartupIntent } from "@/utils/startup-intent" export type SessionTab = { type: "session" @@ -75,6 +76,14 @@ export const { use: useTabs, provider: TabsProvider } = createSimpleContext({ }) const navigateTab = (tab: Tab) => { + // Persist intent so the next launch can pre-warm this session before + // showing the UI (StartupSplashGate reads this synchronously on mount). + writeStartupIntent({ + server: tab.server, + directory: base64Decode(tab.dirBase64), + sessionId: tab.sessionId, + at: Date.now(), + }) setActive("key", tabKey(tab)) const href = tabHref(tab) if (tab.server === server.key) { diff --git a/packages/app/src/utils/startup-intent.ts b/packages/app/src/utils/startup-intent.ts new file mode 100644 index 00000000..d209525f --- /dev/null +++ b/packages/app/src/utils/startup-intent.ts @@ -0,0 +1,69 @@ +/** + * Startup intent snapshot — persists the user's last active session to localStorage + * so the renderer can navigate directly to it on next launch without waiting for + * server-side persisted state to load. + * + * Written synchronously every time the user navigates to a session tab. + * Read synchronously at startup (before any server request). + */ + +const STARTUP_INTENT_KEY = "deepagent:startup-intent" +/** Ignore intents older than 24 h (user may have switched projects externally). */ +const INTENT_MAX_AGE_MS = 24 * 60 * 60 * 1000 + +export type StartupIntent = { + /** ServerConnection.Key — e.g. "local" or "server:" */ + server: string + /** Absolute filesystem path of the project directory */ + directory: string + /** Session ID to restore (may be stale; the session page will handle that gracefully) */ + sessionId: string + /** Unix timestamp (ms) when this intent was written */ + at: number +} + +/** + * Persist the user's current session as the startup intent. + * Synchronous — safe to call on every tab navigation (< 0.1 ms). + */ +export function writeStartupIntent(intent: StartupIntent): void { + try { + localStorage.setItem(STARTUP_INTENT_KEY, JSON.stringify(intent)) + } catch { + // localStorage unavailable (private browsing, quota, etc.) — silently ignore + } +} + +/** + * Read the startup intent written by a previous session. + * Returns null if none exists, it is malformed, or it is older than 24 h. + * Synchronous — safe to call during component initialisation. + */ +export function readStartupIntent(): StartupIntent | null { + try { + const raw = localStorage.getItem(STARTUP_INTENT_KEY) + if (!raw) return null + const parsed = JSON.parse(raw) as Partial + if ( + typeof parsed.server !== "string" || + typeof parsed.directory !== "string" || + typeof parsed.sessionId !== "string" || + typeof parsed.at !== "number" + ) { + return null + } + if (Date.now() - parsed.at > INTENT_MAX_AGE_MS) return null + return parsed as StartupIntent + } catch { + return null + } +} + +/** + * How fresh an intent must be for the StartupSplashGate to honour it for + * immediate navigation (separate from the 24 h existence window). If the app + * was last used > 30 s ago the gate still preloads but uses the tabs fallback + * for the final navigate (avoids navigating to a stale session after e.g. a + * reboot). The 24 h window keeps the intent available for warmup preloading. + */ +export const INTENT_NAVIGATE_WINDOW_MS = 30_000 From 7d02f5d6f80a0821bda063f6307da20999592d39 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 11:01:12 +0800 Subject: [PATCH 08/22] =?UTF-8?q?fix(startup):=20D=20disappears=20?= =?UTF-8?q?=E2=86=92=20direct=20session,=20no=20home=20flash=20=E2=80=94?= =?UTF-8?q?=20real=20fix=20(r6=20v2)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Root cause: StartupSplashGate was inside AppInterface which only mounts AFTER the D splash (controlled by ready() in index.tsx) already disappears. The gate was in the wrong activation point — it ran after the flash, not before. Real fix: prime MemoryHistory before AppInterface mounts. packages/desktop/src/renderer/index.tsx: - Read startup intent synchronously when ready() first becomes true (= server ready, before any component mounts). - Call createMemoryHistory() then history.set({value: sessionPath}) so the router's initial location is the session URL, not '/'. - Pass the pre-seeded history to MemoryRouter via the router prop of AppInterface. - AppInterface now mounts with the router already at the session page; the home route (HomeRoute) is never rendered, eliminating the visible flash. - Inline readLocalStartupIntent + encodeBase64Url to avoid cross-package compiled-artifact resolution issues. packages/app/src/App.tsx (StartupSplashGate): - Add location.pathname !== '/' early return so the gate is a no-op when the router already started at the session URL (avoids spurious one-frame splash). packages/app/src/index.ts: - Export readStartupIntent + INTENT_NAVIGATE_WINDOW_MS (for future use). Co-Authored-By: Claude Opus 4.8 (1M context) --- packages/app/src/app.tsx | 6 ++- packages/app/src/index.ts | 1 + packages/desktop/src/renderer/index.tsx | 59 ++++++++++++++++++++++--- 3 files changed, 58 insertions(+), 8 deletions(-) diff --git a/packages/app/src/app.tsx b/packages/app/src/app.tsx index 73c77365..d736e76b 100644 --- a/packages/app/src/app.tsx +++ b/packages/app/src/app.tsx @@ -164,11 +164,13 @@ function SessionProviders(props: ParentProps) { function StartupSplashGate(props: ParentProps) { const server = useServer() const navigate = useNavigate() + const location = useLocation() const intent = readStartupIntent() - // No usable intent → skip the gate entirely (zero overhead). - if (!intent || intent.server !== server.key) { + // No usable intent, server mismatch, or router already at the session URL + // (index.tsx set MemoryRouter's initialEntries to the session path) → pass through. + if (!intent || intent.server !== server.key || location.pathname !== "/") { return <>{props.children} } diff --git a/packages/app/src/index.ts b/packages/app/src/index.ts index 46bc70e4..65b486a4 100644 --- a/packages/app/src/index.ts +++ b/packages/app/src/index.ts @@ -21,3 +21,4 @@ export { } from "./wsl/types" export { ServerConnection } from "./context/server" export { handleNotificationClick } from "./utils/notification-click" +export { readStartupIntent, INTENT_NAVIGATE_WINDOW_MS } from "./utils/startup-intent" diff --git a/packages/desktop/src/renderer/index.tsx b/packages/desktop/src/renderer/index.tsx index 8ff32622..c1d2633a 100644 --- a/packages/desktop/src/renderer/index.tsx +++ b/packages/desktop/src/renderer/index.tsx @@ -17,7 +17,8 @@ import { import type { UpdaterState } from "@deepagent-code/app/updater" import * as Sentry from "@sentry/solid" import type { AsyncStorage } from "@solid-primitives/storage" -import { MemoryRouter } from "@solidjs/router" +import { MemoryRouter, createMemoryHistory } from "@solidjs/router" +import type { BaseRouterProps } from "@solidjs/router" import { createEffect, createMemo, createResource, createSignal, onCleanup, onMount, Show } from "solid-js" import { render } from "solid-js/web" import pkg from "../../package.json" @@ -29,6 +30,32 @@ import "./styles.css" import { Splash } from "@deepagent-code/ui/logo" import { useTheme } from "@deepagent-code/ui/theme/context" +// Inline startup-intent reader — mirrors packages/app/src/utils/startup-intent.ts. +// Inlined here to avoid cross-package compiled-artifact resolution issues. +const STARTUP_INTENT_KEY = "deepagent:startup-intent" +const INTENT_NAVIGATE_WINDOW_MS = 30_000 + +function readLocalStartupIntent() { + try { + const raw = localStorage.getItem(STARTUP_INTENT_KEY) + if (!raw) return null + const p = JSON.parse(raw) as Record + if ( + typeof p.server !== "string" || + typeof p.directory !== "string" || + typeof p.sessionId !== "string" || + typeof p.at !== "number" + ) return null + if (Date.now() - p.at > INTENT_NAVIGATE_WINDOW_MS) return null + return p as { server: string; directory: string; sessionId: string; at: number } + } catch { return null } +} + +/** URL-safe base64 encode (matches base64Encode from @deepagent-code/core/util/encode). */ +function encodeBase64Url(value: string): string { + return btoa(value).replace(/\+/g, "-").replace(/\//g, "_").replace(/=/g, "") +} + // renderer.initialization — start: renderer module begins executing. // Captured at module level so it includes all synchronous setup before render(). const rendererStartTime = performance.now() @@ -381,11 +408,31 @@ render(() => { return ( - {(key) => ( - - - - )} + {(key) => { + // Read the startup intent synchronously at the moment ready() first becomes + // true. If it is fresh (< INTENT_NAVIGATE_WINDOW_MS) and the server matches, + // prime MemoryHistory to start at the session URL so AppInterface never + // renders the home route — eliminating the visible home-page flash. + const intent = readLocalStartupIntent() + const history = createMemoryHistory() + if (intent && intent.server === key) { + const dirBase64 = encodeBase64Url(intent.directory) + history.set({ + value: `/${dirBase64}/session/${intent.sessionId}`, + replace: true, + }) + } + + const startupRouter = (props: BaseRouterProps) => ( + + ) + + return ( + + + + ) + }} ) From 10f250d580d7824f0a07df591d1188008375aec9 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 11:01:42 +0800 Subject: [PATCH 09/22] fix(sidebar): stabilize For loop keys to prevent mass SortableProject remount MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Root cause of both the 3-second project-switch lag AND the recurring DnD 'Cannot remove nonexistent' errors: layout.projects.list() runs a .map() that creates new object references on every memo invocation. SolidJS For<> tracks items by VALUE IDENTITY (===), so new object references = all items are 'new' = destroy+recreate every SortableProject component on each update: 1. Every SortableWorkspace's serverSync.child(bootstrap:true) fires at once → simultaneous API calls → 3+ second apparent freeze 2. createSortable onCleanup fires for all items while DragDropProvider is still alive but mid-reconciliation → 'Cannot remove nonexistent' DnD errors 3. After enough cycles, stale DnD sensor state blocks subsequent clicks Fix: iterate For<> over worktree STRINGS (stable primitives, same path === same identity) instead of project objects. The project data is looked up reactively inside each slot via createMemo(). SortableProject components now stay alive across icon-color updates and project navigations — only actually added/removed projects cause mount/unmount. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../app/src/pages/layout/sidebar-shell.tsx | 27 ++++++++++++++++++- 1 file changed, 26 insertions(+), 1 deletion(-) diff --git a/packages/app/src/pages/layout/sidebar-shell.tsx b/packages/app/src/pages/layout/sidebar-shell.tsx index b83caf53..429037f7 100644 --- a/packages/app/src/pages/layout/sidebar-shell.tsx +++ b/packages/app/src/pages/layout/sidebar-shell.tsx @@ -75,7 +75,32 @@ export const SidebarContent = (props: {
p.worktree)}> - {(project) => props.renderProject(project)} + {/* + * Iterate over stable worktree STRINGS, not project objects. + * + * layout.projects.list() creates new object references on every + * memo run (via .map({ ...project, icon })). Using those objects + * as For keys causes SolidJS to destroy+recreate EVERY SortableProject + * on each icon-color update or navigation, which: + * 1. Mass-fires DnD cleanup before DragDropProvider is gone → errors + * 2. Triggers every SortableWorkspace's bootstrap API call at once → 3s lag + * + * Worktree strings are stable primitives (same path === same identity). + * The project data updates reactively inside each component via the + * `projectFor` accessor below. + */} + p.worktree)}> + {(worktree) => { + const project = createMemo( + () => props.projects().find((p) => p.worktree === worktree), + ) + return ( + + {(p) => props.renderProject(p)} + + ) + }} + Date: Fri, 24 Jul 2026 13:36:13 +0800 Subject: [PATCH 10/22] fix(desktop): wait for restored session before reveal --- packages/app/src/app.tsx | 176 +++++++++---------- packages/app/src/context/tabs.tsx | 11 +- packages/app/src/index.ts | 1 - packages/app/src/pages/layout.tsx | 7 +- packages/app/src/utils/startup-intent.ts | 69 -------- packages/app/src/utils/startup-ready.test.ts | 48 +++++ packages/app/src/utils/startup-ready.ts | 20 +++ packages/desktop/src/renderer/index.tsx | 127 +++++++------ 8 files changed, 229 insertions(+), 230 deletions(-) delete mode 100644 packages/app/src/utils/startup-intent.ts create mode 100644 packages/app/src/utils/startup-ready.test.ts create mode 100644 packages/app/src/utils/startup-ready.ts diff --git a/packages/app/src/app.tsx b/packages/app/src/app.tsx index d736e76b..57dead0e 100644 --- a/packages/app/src/app.tsx +++ b/packages/app/src/app.tsx @@ -9,7 +9,7 @@ import { Font } from "@deepagent-code/ui/font" import { Splash } from "@deepagent-code/ui/logo" import { ThemeProvider } from "@deepagent-code/ui/theme/context" import { MetaProvider } from "@solidjs/meta" -import { type BaseRouterProps, Navigate, Route, Router, useLocation, useNavigate } from "@solidjs/router" +import { type BaseRouterProps, Navigate, Route, Router, useLocation, useParams } from "@solidjs/router" import { QueryClient, QueryClientProvider } from "@tanstack/solid-query" import { Effect } from "effect" import { @@ -23,7 +23,6 @@ import { type JSX, lazy, onCleanup, - onMount, type ParentProps, Show, } from "solid-js" @@ -35,7 +34,7 @@ import { FileProvider } from "@/context/file" import type { DesktopApi } from "@/utils/desktop-api" import { GatewayProvider } from "@/context/gateway" import { ServerSDKProvider } from "@/context/server-sdk" -import { ServerSyncProvider } from "@/context/server-sync" +import { ServerSyncProvider, useServerSync } from "@/context/server-sync" import { GlobalProvider } from "@/context/global" import { HighlightsProvider } from "@/context/highlights" import { LanguageProvider, type Locale, useLanguage } from "@/context/language" @@ -47,14 +46,13 @@ import { PromptProvider } from "@/context/prompt" import { ServerConnection, ServerProvider, serverName, useServer } from "@/context/server" import { SettingsProvider } from "@/context/settings" import { TerminalProvider } from "@/context/terminal" -import { startupTab, TabsProvider, useTabs } from "@/context/tabs" +import { TabsProvider } from "@/context/tabs" import { WslServersProvider } from "@/wsl/context" -import DirectoryLayout from "@/pages/directory-layout" +import DirectoryLayout, { decodeDirectory } from "@/pages/directory-layout" import Layout from "@/pages/layout" import { ErrorPage } from "./pages/error" import { useCheckServerHealth } from "./utils/server-health" -import { readStartupIntent, INTENT_NAVIGATE_WINDOW_MS } from "@/utils/startup-intent" -import { base64Encode } from "@deepagent-code/core/util/encode" +import { startupViewReady } from "@/utils/startup-ready" const HomeRoute = lazy(() => import("@/pages/home")) const Session = lazy(() => import("@/pages/session")) @@ -108,7 +106,9 @@ function BodyDesignClass() { return null } -function AppShellProviders(props: ParentProps) { +function AppShellProviders(props: ParentProps<{ onStartupReady?: () => void }>) { + const [startupRestoreSettled, setStartupRestoreSettled] = createSignal(false) + return ( @@ -118,7 +118,12 @@ function AppShellProviders(props: ParentProps) { - {props.children} + setStartupRestoreSettled(true)}> + {props.onStartupReady ? ( + + ) : null} + {props.children} + @@ -143,100 +148,74 @@ function SessionProviders(props: ParentProps) { ) } -/** - * StartupSplashGate — keeps the D splash visible after ConnectionGate passes and - * navigates directly to the last session before revealing the app shell. - * - * Must be rendered inside a router root (needs useNavigate) and useServer. - * Reads the startup-intent snapshot written synchronously by navigateTab (tabs.tsx). - * - * Gate logic: - * • No intent / server mismatch → pass-through immediately (new install, cleared state). - * • Fresh intent (< INTENT_NAVIGATE_WINDOW_MS) → navigate to last session, release. - * • Stale intent (≥ window but < 24 h) → release immediately; the existing - * tabs.ready() effect in RouterRoot handles navigation as before. - * - * NOTE: The gate does NOT attempt to pre-warm the session query cache because the - * session SDK responses are keyed inside ServerSyncProvider (query key: [scope, dir, - * 'loadSessions']) which is not yet mounted when StartupSplashGate's onMount fires. - * The value of the gate is eliminating the visible home-page flash, not pre-fetching. - */ -function StartupSplashGate(props: ParentProps) { +function StartupViewReady(props: { restoreSettled: boolean; onReady: () => void }) { const server = useServer() - const navigate = useNavigate() + const serverSync = useServerSync() const location = useLocation() + const params = useParams() + let complete = false + let signature = "" - const intent = readStartupIntent() - - // No usable intent, server mismatch, or router already at the session URL - // (index.tsx set MemoryRouter's initialEntries to the session path) → pass through. - if (!intent || intent.server !== server.key || location.pathname !== "/") { - return <>{props.children} + const state = () => { + const directory = params.dir ? decodeDirectory(params.dir) : undefined + const sessionId = params.id + const store = directory ? serverSync.peek(directory, { bootstrap: false })[0] : undefined + return { + pathname: location.pathname, + serverReady: server.ready(), + globalReady: serverSync.ready, + globalError: !!serverSync.error, + restoreSettled: props.restoreSettled, + lastProject: server.projects.last(), + directory, + directoryReady: !directory || (!!store && store.status !== "loading"), + sessionId, + hasSession: !!store?.session.some((session) => session.id === sessionId), + messagesReady: !!sessionId && store?.message[sessionId] !== undefined, + } } - const [warmupDone, setWarmupDone] = createSignal(false) - - onMount(() => { - try { - // Navigate immediately when the intent is fresh (process crash / hot restart). - // Stale intents delegate to the tabs.ready() effect so a long-idle cold-start - // doesn't force navigation to a session the user may no longer want. - if (Date.now() - intent.at < INTENT_NAVIGATE_WINDOW_MS) { - const dirBase64 = base64Encode(intent.directory) - navigate(`/${dirBase64}/session/${intent.sessionId}`, { replace: true }) - } - } finally { - setWarmupDone(true) + createEffect(() => { + const current = state() + const nextSignature = JSON.stringify({ + route: current.pathname === "/" ? "home" : current.sessionId ? "session" : "directory", + serverReady: current.serverReady, + globalReady: current.globalReady, + globalError: current.globalError, + restoreSettled: current.restoreSettled, + hasLastProject: !!current.lastProject, + hasDirectory: !!current.directory, + directoryReady: current.directoryReady, + hasSessionId: !!current.sessionId, + hasSession: current.hasSession, + messagesReady: current.messagesReady, + }) + if (signature !== nextSignature) { + signature = nextSignature + console.info("[startup] readiness", nextSignature) } + + if (complete || !startupViewReady(current)) return + complete = true + props.onReady() }) - return ( - - -
- } - > - {props.children} - - ) + return null } -function RouterRoot(props: ParentProps<{ appChildren?: JSX.Element }>) { - const tabs = useTabs() - const server = useServer() - const navigate = useNavigate() - const location = useLocation() - - // Restore the explicitly persisted active tab. A cross-server directory is never - // navigated until the target server has become active. - // Skip when StartupSplashGate has a fresh intent — it navigates directly and this - // effect would race/conflict with it during the warmup window. - createEffect(() => { - if (!tabs.ready()) return - if (location.pathname !== "/") return - const intent = readStartupIntent() - if (intent && intent.server === server.key && Date.now() - intent.at < INTENT_NAVIGATE_WINDOW_MS) return - const tab = startupTab(tabs.store, tabs.active.key, server.list) - if (!tab) return - if (server.key !== tab.server) { - server.setActive(tab.server) - return - } - navigate(`/${tab.dirBase64}/session`, { replace: true }) - }) - +function RouterRoot( + props: ParentProps<{ + appChildren?: JSX.Element + onStartupReady?: () => void + }>, +) { return ( - - - {/*}>*/} - {props.appChildren} - {props.children} - {/**/} - - + + {/*}>*/} + {props.appChildren} + {props.children} + {/**/} + ) } @@ -397,6 +376,7 @@ export function AppInterface(props: { servers?: Array router?: Component disableHealthCheck?: boolean + onStartupReady?: () => void }) { return ( @@ -415,7 +395,9 @@ export function AppInterface(props: { - {routerProps.children} + + {routerProps.children} + @@ -425,10 +407,10 @@ export function AppInterface(props: { > - } /> - - - + } /> + + + diff --git a/packages/app/src/context/tabs.tsx b/packages/app/src/context/tabs.tsx index 6ef98899..64ac3e46 100644 --- a/packages/app/src/context/tabs.tsx +++ b/packages/app/src/context/tabs.tsx @@ -1,13 +1,12 @@ import type { Session } from "@deepagent-code/sdk/v2/client" import { createSimpleContext } from "@deepagent-code/ui/context" -import { base64Encode, base64Decode } from "@deepagent-code/core/util/encode" +import { base64Encode } from "@deepagent-code/core/util/encode" import { createStore, produce } from "solid-js/store" import { Persist, persisted } from "@/utils/persist" import { ServerConnection, useServer } from "./server" import { createEffect, startTransition } from "solid-js" import { useNavigate, useParams } from "@solidjs/router" import { SessionTabsRemovedDetail } from "@/components/titlebar-session-events" -import { writeStartupIntent } from "@/utils/startup-intent" export type SessionTab = { type: "session" @@ -76,14 +75,6 @@ export const { use: useTabs, provider: TabsProvider } = createSimpleContext({ }) const navigateTab = (tab: Tab) => { - // Persist intent so the next launch can pre-warm this session before - // showing the UI (StartupSplashGate reads this synchronously on mount). - writeStartupIntent({ - server: tab.server, - directory: base64Decode(tab.dirBase64), - sessionId: tab.sessionId, - at: Date.now(), - }) setActive("key", tabKey(tab)) const href = tabHref(tab) if (tab.server === server.key) { diff --git a/packages/app/src/index.ts b/packages/app/src/index.ts index 65b486a4..46bc70e4 100644 --- a/packages/app/src/index.ts +++ b/packages/app/src/index.ts @@ -21,4 +21,3 @@ export { } from "./wsl/types" export { ServerConnection } from "./context/server" export { handleNotificationClick } from "./utils/notification-click" -export { readStartupIntent, INTENT_NAVIGATE_WINDOW_MS } from "./utils/startup-intent" diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index 42a325f1..55705fdb 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -95,7 +95,7 @@ import { import { ProjectDragOverlay, SortableProject, type ProjectSidebarContext } from "./layout/sidebar-project" import { SidebarContent } from "./layout/sidebar-shell" -export default function Layout(props: ParentProps) { +export default function Layout(props: ParentProps<{ onStartupRestoreSettled?: () => void }>) { const serverSDK = useServerSDK() const [store, setStore, , ready] = persisted( Persist.serverGlobal(serverSDK.scope, "layout.page", ["layout.page.v1"]), @@ -601,6 +601,11 @@ export default function Layout(props: ParentProps) { } }) + createEffect(() => { + if (autoselecting.loading) return + props.onStartupRestoreSettled?.() + }) + const workspaceName = (directory: string, projectId?: string, branch?: string) => { const key = pathKey(directory) const direct = store.workspaceName[key] ?? store.workspaceName[directory] diff --git a/packages/app/src/utils/startup-intent.ts b/packages/app/src/utils/startup-intent.ts deleted file mode 100644 index d209525f..00000000 --- a/packages/app/src/utils/startup-intent.ts +++ /dev/null @@ -1,69 +0,0 @@ -/** - * Startup intent snapshot — persists the user's last active session to localStorage - * so the renderer can navigate directly to it on next launch without waiting for - * server-side persisted state to load. - * - * Written synchronously every time the user navigates to a session tab. - * Read synchronously at startup (before any server request). - */ - -const STARTUP_INTENT_KEY = "deepagent:startup-intent" -/** Ignore intents older than 24 h (user may have switched projects externally). */ -const INTENT_MAX_AGE_MS = 24 * 60 * 60 * 1000 - -export type StartupIntent = { - /** ServerConnection.Key — e.g. "local" or "server:" */ - server: string - /** Absolute filesystem path of the project directory */ - directory: string - /** Session ID to restore (may be stale; the session page will handle that gracefully) */ - sessionId: string - /** Unix timestamp (ms) when this intent was written */ - at: number -} - -/** - * Persist the user's current session as the startup intent. - * Synchronous — safe to call on every tab navigation (< 0.1 ms). - */ -export function writeStartupIntent(intent: StartupIntent): void { - try { - localStorage.setItem(STARTUP_INTENT_KEY, JSON.stringify(intent)) - } catch { - // localStorage unavailable (private browsing, quota, etc.) — silently ignore - } -} - -/** - * Read the startup intent written by a previous session. - * Returns null if none exists, it is malformed, or it is older than 24 h. - * Synchronous — safe to call during component initialisation. - */ -export function readStartupIntent(): StartupIntent | null { - try { - const raw = localStorage.getItem(STARTUP_INTENT_KEY) - if (!raw) return null - const parsed = JSON.parse(raw) as Partial - if ( - typeof parsed.server !== "string" || - typeof parsed.directory !== "string" || - typeof parsed.sessionId !== "string" || - typeof parsed.at !== "number" - ) { - return null - } - if (Date.now() - parsed.at > INTENT_MAX_AGE_MS) return null - return parsed as StartupIntent - } catch { - return null - } -} - -/** - * How fresh an intent must be for the StartupSplashGate to honour it for - * immediate navigation (separate from the 24 h existence window). If the app - * was last used > 30 s ago the gate still preloads but uses the tabs fallback - * for the final navigate (avoids navigating to a stale session after e.g. a - * reboot). The 24 h window keeps the intent available for warmup preloading. - */ -export const INTENT_NAVIGATE_WINDOW_MS = 30_000 diff --git a/packages/app/src/utils/startup-ready.test.ts b/packages/app/src/utils/startup-ready.test.ts new file mode 100644 index 00000000..d31bb59a --- /dev/null +++ b/packages/app/src/utils/startup-ready.test.ts @@ -0,0 +1,48 @@ +import { describe, expect, test } from "bun:test" +import { startupViewReady } from "./startup-ready" + +const ready = { + pathname: "/project/session/session-1", + serverReady: true, + globalReady: true, + globalError: false, + restoreSettled: true, + directory: "/workspace/project", + directoryReady: true, + sessionId: "session-1", + hasSession: true, + messagesReady: true, +} + +describe("startup view readiness", () => { + test("keeps the home route covered while the persisted project is restoring", () => { + expect( + startupViewReady({ + ...ready, + pathname: "/", + directory: undefined, + sessionId: undefined, + restoreSettled: false, + }), + ).toBe(false) + expect(startupViewReady({ ...ready, pathname: "/", directory: undefined, sessionId: undefined })).toBe(true) + }) + + test("waits for global and directory synchronization", () => { + expect(startupViewReady({ ...ready, serverReady: false })).toBe(false) + expect(startupViewReady({ ...ready, globalReady: false })).toBe(false) + expect(startupViewReady({ ...ready, restoreSettled: false })).toBe(false) + expect(startupViewReady({ ...ready, directoryReady: false })).toBe(false) + }) + + test("requires restored session metadata and initial messages", () => { + expect(startupViewReady(ready)).toBe(true) + expect(startupViewReady({ ...ready, hasSession: false })).toBe(false) + expect(startupViewReady({ ...ready, messagesReady: false })).toBe(false) + }) + + test("allows a synchronized new-session route and server error state", () => { + expect(startupViewReady({ ...ready, sessionId: undefined })).toBe(true) + expect(startupViewReady({ ...ready, globalReady: false, globalError: true })).toBe(true) + }) +}) diff --git a/packages/app/src/utils/startup-ready.ts b/packages/app/src/utils/startup-ready.ts new file mode 100644 index 00000000..24fb7569 --- /dev/null +++ b/packages/app/src/utils/startup-ready.ts @@ -0,0 +1,20 @@ +export function startupViewReady(input: { + pathname: string + serverReady: boolean + globalReady: boolean + globalError: boolean + restoreSettled: boolean + directory?: string + directoryReady: boolean + sessionId?: string + hasSession: boolean + messagesReady: boolean +}) { + if (input.globalError) return true + if (!input.serverReady || !input.globalReady || !input.restoreSettled) return false + if (input.pathname === "/") return true + if (!input.directory) return true + if (!input.directoryReady) return false + if (!input.sessionId) return true + return input.hasSession && input.messagesReady +} diff --git a/packages/desktop/src/renderer/index.tsx b/packages/desktop/src/renderer/index.tsx index c1d2633a..33b95d1d 100644 --- a/packages/desktop/src/renderer/index.tsx +++ b/packages/desktop/src/renderer/index.tsx @@ -17,8 +17,7 @@ import { import type { UpdaterState } from "@deepagent-code/app/updater" import * as Sentry from "@sentry/solid" import type { AsyncStorage } from "@solid-primitives/storage" -import { MemoryRouter, createMemoryHistory } from "@solidjs/router" -import type { BaseRouterProps } from "@solidjs/router" +import { type BaseRouterProps, MemoryRouter, createMemoryHistory } from "@solidjs/router" import { createEffect, createMemo, createResource, createSignal, onCleanup, onMount, Show } from "solid-js" import { render } from "solid-js/web" import pkg from "../../package.json" @@ -30,31 +29,8 @@ import "./styles.css" import { Splash } from "@deepagent-code/ui/logo" import { useTheme } from "@deepagent-code/ui/theme/context" -// Inline startup-intent reader — mirrors packages/app/src/utils/startup-intent.ts. -// Inlined here to avoid cross-package compiled-artifact resolution issues. -const STARTUP_INTENT_KEY = "deepagent:startup-intent" -const INTENT_NAVIGATE_WINDOW_MS = 30_000 - -function readLocalStartupIntent() { - try { - const raw = localStorage.getItem(STARTUP_INTENT_KEY) - if (!raw) return null - const p = JSON.parse(raw) as Record - if ( - typeof p.server !== "string" || - typeof p.directory !== "string" || - typeof p.sessionId !== "string" || - typeof p.at !== "number" - ) return null - if (Date.now() - p.at > INTENT_NAVIGATE_WINDOW_MS) return null - return p as { server: string; directory: string; sessionId: string; at: number } - } catch { return null } -} - -/** URL-safe base64 encode (matches base64Encode from @deepagent-code/core/util/encode). */ -function encodeBase64Url(value: string): string { - return btoa(value).replace(/\+/g, "-").replace(/\//g, "_").replace(/=/g, "") -} +const STARTUP_SLOW_MS = 15_000 +const STARTUP_HARD_TIMEOUT_MS = 120_000 // renderer.initialization — start: renderer module begins executing. // Captured at module level so it includes all synchronous setup before render(). @@ -405,34 +381,81 @@ render(() => { ServerConnection.Key.make(availableStartupServer(defaultServer.latest, wslServers.data)), ) + function RoutedApp(props: { serverKey: ServerConnection.Key }) { + const history = createMemoryHistory() + const startupRouter = (props: BaseRouterProps) => + const [startupState, setStartupState] = createSignal<"waiting" | "ready" | "timeout">("waiting") + const startupStartedAt = performance.now() + let slowTimeout: number | undefined + let hardTimeout: number | undefined + + const finishStartup = (result: "ready" | "timeout") => { + if (startupState() !== "waiting") return + if (slowTimeout !== undefined) window.clearTimeout(slowTimeout) + if (hardTimeout !== undefined) window.clearTimeout(hardTimeout) + slowTimeout = undefined + hardTimeout = undefined + const payload = { + event: "startup.app_reveal", + result, + durationMs: Math.round(performance.now() - startupStartedAt), + } + if (result === "timeout") console.warn("[startup] telemetry", JSON.stringify(payload)) + if (result === "ready") console.info("[startup] telemetry", JSON.stringify(payload)) + setStartupState(result) + } + + onMount(() => { + slowTimeout = window.setTimeout(() => { + if (startupState() !== "waiting") return + console.warn( + "[startup] telemetry", + JSON.stringify({ + event: "startup.app_reveal", + result: "slow", + durationMs: Math.round(performance.now() - startupStartedAt), + }), + ) + }, STARTUP_SLOW_MS) + hardTimeout = window.setTimeout(() => finishStartup("timeout"), STARTUP_HARD_TIMEOUT_MS) + onCleanup(() => { + if (slowTimeout !== undefined) window.clearTimeout(slowTimeout) + if (hardTimeout !== undefined) window.clearTimeout(hardTimeout) + }) + }) + + const waiting = () => startupState() === "waiting" + + return ( + <> +
+ finishStartup("ready")} + > + + +
+ +
+ +
+ + + ) + } + return ( - {(key) => { - // Read the startup intent synchronously at the moment ready() first becomes - // true. If it is fresh (< INTENT_NAVIGATE_WINDOW_MS) and the server matches, - // prime MemoryHistory to start at the session URL so AppInterface never - // renders the home route — eliminating the visible home-page flash. - const intent = readLocalStartupIntent() - const history = createMemoryHistory() - if (intent && intent.server === key) { - const dirBase64 = encodeBase64Url(intent.directory) - history.set({ - value: `/${dirBase64}/session/${intent.sessionId}`, - replace: true, - }) - } - - const startupRouter = (props: BaseRouterProps) => ( - - ) - - return ( - - - - ) - }} + {(key) => } ) From 18744862a9a4153f6dd7743a8546751cf77b1290 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Fri, 24 Jul 2026 16:16:54 +0800 Subject: [PATCH 11/22] fix(session): contain unfinished runner and define prompt epochs --- .../test/server/httpapi-session.test.ts | 25 +++++++++++++++++-- packages/server/src/groups/session.ts | 5 ++-- packages/server/src/handlers.ts | 2 -- packages/server/src/handlers/session.ts | 18 ++++++++++++- 4 files changed, 43 insertions(+), 7 deletions(-) diff --git a/packages/deepagent-code/test/server/httpapi-session.test.ts b/packages/deepagent-code/test/server/httpapi-session.test.ts index 70b9b6b9..aa685aef 100644 --- a/packages/deepagent-code/test/server/httpapi-session.test.ts +++ b/packages/deepagent-code/test/server/httpapi-session.test.ts @@ -606,7 +606,7 @@ describe("session HttpApi", () => { request(`/api/session/${session.id}/prompt`, { method: "POST", headers: { ...headers, "content-type": "application/json" }, - body: JSON.stringify({ id: "msg_http_prompt", prompt: { text: "hello" } }), + body: JSON.stringify({ id: "msg_http_prompt", prompt: { text: "hello" }, resume: false }), }) const first = yield* recordPrompt() const retried = yield* recordPrompt() @@ -641,7 +641,7 @@ describe("session HttpApi", () => { const conflict = yield* request(`/api/session/${session.id}/prompt`, { method: "POST", headers: { ...headers, "content-type": "application/json" }, - body: JSON.stringify({ id: "msg_http_prompt", prompt: { text: "goodbye" } }), + body: JSON.stringify({ id: "msg_http_prompt", prompt: { text: "goodbye" }, resume: false }), }) expect(conflict.status).toBe(409) expect(yield* responseJson(conflict)).toEqual({ @@ -676,6 +676,27 @@ describe("session HttpApi", () => { message: "Session wait is not available yet", service: "session.wait", }) + + const prompt = yield* request(`/api/session/${session.id}/prompt`, { + method: "POST", + headers: { ...headers, "content-type": "application/json" }, + body: JSON.stringify({ id: "msg_execution_unavailable", prompt: { text: "hello" } }), + }) + expect(prompt.status).toBe(503) + expect(yield* responseJson(prompt)).toEqual({ + _tag: "ServiceUnavailableError", + message: "Session execution is not available on this endpoint", + service: "session.prompt", + }) + const admitted = yield* Database.Service.use(({ db }) => + db + .select() + .from(SessionInputTable) + .where(eq(SessionInputTable.id, SessionMessage.ID.make("msg_execution_unavailable"))) + .get() + .pipe(Effect.orDie), + ) + expect(admitted).toBeUndefined() }), { git: true, config: { formatter: false, lsp: false } }, ) diff --git a/packages/server/src/groups/session.ts b/packages/server/src/groups/session.ts index 8f0f4cff..b79d85fa 100644 --- a/packages/server/src/groups/session.ts +++ b/packages/server/src/groups/session.ts @@ -115,14 +115,15 @@ export const SessionGroup = HttpApiGroup.make("server.session") resume: Schema.Boolean.pipe(Schema.optional), }), success: Schema.Struct({ data: SessionInput.Admitted }), - error: [ConflictError, SessionNotFoundError], + error: [ConflictError, ServiceUnavailableError, SessionNotFoundError], }) .middleware(SessionLocationMiddleware) .annotateMerge( OpenApi.annotations({ identifier: "v2.session.prompt", summary: "Send message", - description: "Durably admit one session input and schedule agent-loop execution unless resume is false.", + description: + "Durably admit one session input when resume is false. This endpoint does not execute the production SessionPrompt engine.", }), ), ) diff --git a/packages/server/src/handlers.ts b/packages/server/src/handlers.ts index 50f7bc2f..73d1eb09 100644 --- a/packages/server/src/handlers.ts +++ b/packages/server/src/handlers.ts @@ -16,7 +16,6 @@ import { EventHandler } from "./handlers/event" import { AgentHandler } from "./handlers/agent" import { HealthHandler } from "./handlers/health" import { QuestionHandler } from "./handlers/question" -import * as SessionExecutionLocal from "@deepagent-code/core/session/execution/local" export const handlers = Layer.mergeAll( HealthHandler, @@ -35,7 +34,6 @@ export const handlers = Layer.mergeAll( Layer.provide(sessionLocationLayer), Layer.provide(locationLayer), Layer.provide(SessionV2.defaultLayer), - Layer.provide(SessionExecutionLocal.defaultLayer), Layer.provide(PermissionSaved.defaultLayer), Layer.provide(LocationServiceMap.layer), ) diff --git a/packages/server/src/handlers/session.ts b/packages/server/src/handlers/session.ts index a8ae9baf..4ce72e8a 100644 --- a/packages/server/src/handlers/session.ts +++ b/packages/server/src/handlers/session.ts @@ -64,6 +64,22 @@ export const SessionHandler = HttpApiBuilder.group(Api, "server.session", (handl .handle( "session.prompt", Effect.fn(function* (ctx) { + if (ctx.payload.resume !== false) { + yield* session.get(ctx.params.sessionID).pipe( + Effect.catchTag("Session.NotFoundError", (error) => + Effect.fail( + new SessionNotFoundError({ + sessionID: error.sessionID, + message: `Session not found: ${error.sessionID}`, + }), + ), + ), + ) + return yield* new ServiceUnavailableError({ + message: "Session execution is not available on this endpoint", + service: "session.prompt", + }) + } return { data: yield* session .prompt({ @@ -71,7 +87,7 @@ export const SessionHandler = HttpApiBuilder.group(Api, "server.session", (handl id: ctx.payload.id, prompt: ctx.payload.prompt, delivery: ctx.payload.delivery, - resume: ctx.payload.resume, + resume: false, }) .pipe( Effect.catchTag("Session.NotFoundError", (error) => From fa43d2bc24f7f530c5cce4cce162bfcb0d7aad91 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:00:56 +0800 Subject: [PATCH 12/22] fix(core): keep knowledge review state live --- .../review/dialog-review-contract.test.ts | 8 ++- .../components/review/dialog-review.api.ts | 8 +++ packages/app/src/pages/layout.tsx | 5 +- packages/core/src/agent-gateway.ts | 41 +++++++----- packages/core/src/deepagent/document-store.ts | 6 ++ .../deepagent/environment-fact-adoption.ts | 18 ++--- .../core/src/deepagent/knowledge-source.ts | 28 ++++++-- .../deepagent/knowledge-source-cache.test.ts | 67 +++++++++++++++++++ .../src/import/writer/memory.ts | 21 +++--- .../instance/httpapi/groups/deepagent.ts | 20 +++++- .../instance/httpapi/handlers/deepagent.ts | 40 ++++++----- .../deepagent/knowledge-import-cache.test.ts | 65 ++++++++++++++++++ 12 files changed, 261 insertions(+), 66 deletions(-) create mode 100644 packages/core/test/deepagent/knowledge-source-cache.test.ts create mode 100644 packages/deepagent-code/test/deepagent/knowledge-import-cache.test.ts diff --git a/packages/app/src/components/review/dialog-review-contract.test.ts b/packages/app/src/components/review/dialog-review-contract.test.ts index 6030a426..0a3b509b 100644 --- a/packages/app/src/components/review/dialog-review-contract.test.ts +++ b/packages/app/src/components/review/dialog-review-contract.test.ts @@ -1,5 +1,5 @@ import { describe, expect, test } from "bun:test" -import { listPending, setStatus, listEnvFacts, decideEnvFact, modifyEnvFact } from "./dialog-review.api" +import { listPending, reviewSummary, setStatus, listEnvFacts, decideEnvFact, modifyEnvFact } from "./dialog-review.api" // P1-C route contract: the V3.1 self-learning Review dialog talks to the raw-request escape-hatch // routes (NOT the generated SDK). These assertions lock the exact method/url/body so a backend @@ -43,6 +43,12 @@ describe("DeepAgent review dialog route contract", () => { expect(await listPending(client(calls, {}))).toEqual([]) }) + test("reviewSummary GETs the lightweight summary route", async () => { + const calls: Recorded[] = [] + expect(await reviewSummary(client(calls, { pendingCount: 3 }))).toEqual({ pendingCount: 3 }) + expect(calls).toEqual([{ method: "GET", url: "/deepagent/knowledge/review-summary" }]) + }) + test("approve POSTs /deepagent/knowledge/approve with { ids }", async () => { const calls: Recorded[] = [] await setStatus(client(calls, { updated: ["a"] }), "approve", ["a", "b"]) diff --git a/packages/app/src/components/review/dialog-review.api.ts b/packages/app/src/components/review/dialog-review.api.ts index 3f293b78..169ef00f 100644 --- a/packages/app/src/components/review/dialog-review.api.ts +++ b/packages/app/src/components/review/dialog-review.api.ts @@ -36,6 +36,14 @@ export const listPending = async (client: ReviewClient): Promise => { + const response = await client.client.request<{ pendingCount?: number }>({ + method: "GET", + url: "/deepagent/knowledge/review-summary", + }) + return { pendingCount: response.data?.pendingCount ?? 0 } +} + export const setStatus = async ( client: ReviewClient, action: "approve" | "reject-ids", diff --git a/packages/app/src/pages/layout.tsx b/packages/app/src/pages/layout.tsx index 55705fdb..c2af49df 100644 --- a/packages/app/src/pages/layout.tsx +++ b/packages/app/src/pages/layout.tsx @@ -63,7 +63,7 @@ import { useTheme, type ColorScheme } from "@deepagent-code/ui/theme/context" import { useCommand, type CommandOption } from "@/context/command" import { ConstrainDragXAxis, getDraggableId, FixedDragDropSensors } from "@/utils/solid-dnd" import { DebugBar } from "@/components/debug-bar" -import { listPending } from "@/components/review/dialog-review.api" +import { reviewSummary } from "@/components/review/dialog-review.api" import { fetchCapabilities } from "@/components/deepagent/panel-goal.api" import { Titlebar, type TitlebarUpdate } from "@/components/titlebar" import { useDirectoryPicker } from "@/components/directory-picker" @@ -167,8 +167,7 @@ export default function Layout(props: ParentProps<{ onStartupRestoreSettled?: () if (!dir) return false try { const sdk = serverSDK.createDirSdkContext(dir) - const items = await listPending(sdk.client as never) - return items.some((item) => item.approval_status === "pending") + return (await reviewSummary(sdk.client as never)).pendingCount > 0 } catch { return false } diff --git a/packages/core/src/agent-gateway.ts b/packages/core/src/agent-gateway.ts index eab8efdb..12d3fc65 100644 --- a/packages/core/src/agent-gateway.ts +++ b/packages/core/src/agent-gateway.ts @@ -262,6 +262,8 @@ export const selfLearningPolicy = (): SelfLearningPolicy => current.selfLearning // No disk writes — available only for diagnostics / budget tracking in-process. type GeneralAuditEntry = { turnCount: number; totalInputTokens: number; totalOutputTokens: number } const generalAuditState = new Map() +const seededKnowledgeBases = new Set() +let configuredStorageBaseDir: string | null = null const recordGeneralAudit = (sessionID: string, inputTokens: number, outputTokens: number): void => { const existing = generalAuditState.get(sessionID) @@ -286,23 +288,27 @@ export const configure = (config: Config = {}) => { // is still observed. This replaces the old path.dirname(runsDir) inference that made durable // knowledge/state diverge from project-memory whenever runsDir pointed outside /runs. const baseDir = config.baseDir ?? resolveDeepAgentCodeHome() - DeepAgentSessionState.configure(path.join(baseDir, "state")) - // I33-1: the structural plan lives in the single DocumentStore under /state/goal//graph - // (co-located with the goal/run graph so the `plan` tool and the goal path write the SAME doc). Root - // it from the SAME baseDir/state session-state uses, so the two paths converge (planStoreRoot ≡ - // goalStoreRoot). Must be set before any plan read/write. - DeepAgentPlanStore.configureRoot(path.join(baseDir, "state")) - // docs/34 §7.2/§8: durable knowledge stores root under baseDir (public/knowledge + project// - // knowledge). The retriever's knowledge-source reads from here. Same injected baseDir, no self-resolve. - DeepAgentKnowledgeSource.configure(baseDir) - // docs/34 §9 DAP-11: seed core in-code knowledge (CORE_STRATEGIES/METHODOLOGY_REGISTRY/gpu pack) - // into the user-global DocumentStore on every configure() call. seedCoreKnowledge is idempotent - // (skips docs that already exist), so re-calling on restart is safe. After seeding, the retriever - // reads these from DocumentStore and the in-code constants are no longer in the retrieval path. - try { - DeepAgentKnowledgeSeed.seedCoreKnowledgeAt(baseDir) - } catch { - /* non-fatal: stale seed, retry next call */ + const resolvedBaseDir = path.resolve(baseDir) + if (configuredStorageBaseDir !== resolvedBaseDir) { + DeepAgentSessionState.configure(path.join(resolvedBaseDir, "state")) + // I33-1: the structural plan lives under /state/goal//graph. Root it from the SAME + // state directory before any plan read/write. SessionState.configure also applies this root; the + // explicit call documents the gateway ownership boundary. + DeepAgentPlanStore.configureRoot(path.join(resolvedBaseDir, "state")) + configuredStorageBaseDir = resolvedBaseDir + } + // configure() is identity-preserving for the same root, so request-time policy updates do not drop + // the live knowledge stores. It also restores the adapter if an explicit test reset it. + DeepAgentKnowledgeSource.configure(resolvedBaseDir) + // Seed each storage root once per process. Reconfiguration changes runtime policy frequently, but + // the built-in corpus only changes across process/app versions and must not rebuild its disk index. + if (!seededKnowledgeBases.has(resolvedBaseDir)) { + try { + DeepAgentKnowledgeSeed.seedCoreKnowledge(DeepAgentKnowledgeSource.userGlobalStoreFor()) + seededKnowledgeBases.add(resolvedBaseDir) + } catch { + /* non-fatal: stale seed, retry next call */ + } } // docs/34 §3: domain pack registry. Built-in packs (packages/domain-packs, bundled with the app) // are ALWAYS discovered automatically. A user/org pack dir can be layered on top via config.packDir @@ -835,7 +841,6 @@ const markStaleEnvironmentFactsFromRun = ( if (stale.length === 0) return const userGlobal = DeepAgentKnowledgeSource.userGlobalStoreFor() for (const id of stale) userGlobal.markEnvironmentFactStale(id) - DeepAgentKnowledgeSource.invalidateCache() } // Concatenate the output of every FAILED validation this run recorded — that's where a tool's diff --git a/packages/core/src/deepagent/document-store.ts b/packages/core/src/deepagent/document-store.ts index e66eee41..1bceab24 100644 --- a/packages/core/src/deepagent/document-store.ts +++ b/packages/core/src/deepagent/document-store.ts @@ -674,6 +674,12 @@ export class DocumentStore { } private findLogical(input: CreateDocInput): Doc | null { const domain = input.domain ?? null + if (input.idSlug) { + const slug = slugify(input.idSlug) + const id = domain ? `doc:${input.type}:${domain}:${slug}` : `doc:${input.type}:${slug}` + const direct = this.get(id) + if (direct?.domain === domain && direct.description === input.description) return direct + } for (const ref of this.list({ type: input.type, scope: input.scope })) { const doc = this.get(ref.id) if (!doc) continue diff --git a/packages/core/src/deepagent/environment-fact-adoption.ts b/packages/core/src/deepagent/environment-fact-adoption.ts index f692a1e7..00cb146a 100644 --- a/packages/core/src/deepagent/environment-fact-adoption.ts +++ b/packages/core/src/deepagent/environment-fact-adoption.ts @@ -3,11 +3,7 @@ import path from "node:path" import type { ProjectPaths } from "./workspace" import type { AdoptionRecord } from "./environment-fact" import { useGateAction, type EnvironmentFactBody } from "./environment-fact" -import { - openUserGlobalStore, - openProjectStore, - type DurableKnowledgeStore, -} from "./durable-knowledge-store" +import { openUserGlobalStore, openProjectStore, type DurableKnowledgeStore } from "./durable-knowledge-store" import type { DocRef } from "./document-store" // V3.8.1 §G.5 use-gate persistence. A project's stance toward each user-global provisional @@ -69,9 +65,10 @@ export class EnvironmentFactAdoption { private readonly baseDir: string, private readonly paths: ProjectPaths, private readonly workspacePath: string, + stores?: { readonly userGlobal: DurableKnowledgeStore; readonly project: DurableKnowledgeStore }, ) { - this.userGlobal = openUserGlobalStore(baseDir) - this.project = openProjectStore(baseDir, workspacePath) + this.userGlobal = stores?.userGlobal ?? openUserGlobalStore(baseDir) + this.project = stores?.project ?? openProjectStore(baseDir, workspacePath) } private records(): AdoptionRecord[] { @@ -172,7 +169,12 @@ export class EnvironmentFactAdoption { domain: input.domain ?? null, provenance: { source: "human" }, }) - this.writeRecord({ fact_id: updated.id, stance: "adopted", decided_at: input.now, adopted_version: updated.version }) + this.writeRecord({ + fact_id: updated.id, + stance: "adopted", + decided_at: input.now, + adopted_version: updated.version, + }) return { updatedId: updated.id } } diff --git a/packages/core/src/deepagent/knowledge-source.ts b/packages/core/src/deepagent/knowledge-source.ts index 2bf841a2..49ade60d 100644 --- a/packages/core/src/deepagent/knowledge-source.ts +++ b/packages/core/src/deepagent/knowledge-source.ts @@ -9,15 +9,16 @@ import { import type { DocType, DocumentStore } from "./document-store" import { DeepAgentCodeHome } from "./workspace" import { EnvironmentFactAdoption } from "./environment-fact-adoption" +import path from "node:path" // V3.2.1 decision B (docs/34 §8): the read-side adapter between the knowledge retriever and the // durable DocumentStore. Durable knowledge lives in TWO roots under the single injected base // (Global.Path.agent.data): user-global (public/knowledge, visible everywhere) and per-project // (project//knowledge, project-shared isolation). A retrieval for a workspace UNIONS both. // -// Mirrors the old memory-store module pattern: a single configured base + a clearable cache, so the -// retriever stays a pure-ish function (retrieve(input)) and approve/reject is reflected after -// invalidateCache(). This module is the ONLY durable read path the retriever uses. +// The configured stores are process-long-lived. All in-process writers use these same handles, so +// changes are immediately visible without rebuilding the disk index. invalidateCache() is reserved +// for explicit cold-reload/testing paths. This module is the ONLY durable read path the retriever uses. let baseDir: string | null = null let userGlobalCache: DurableKnowledgeStore | null = null @@ -31,14 +32,19 @@ let sharedDocumentStore: DocumentStore | null = null // configure, from the injected baseDir — never a self-resolved home). // H32-1: optional sharedStore accepted; passed through to openUserGlobalStore/openProjectStore. export const configure = (dir: string, sharedStore?: DocumentStore): void => { - baseDir = dir - sharedDocumentStore = sharedStore ?? null + const nextBaseDir = path.resolve(dir) + const nextSharedDocumentStore = sharedStore ?? null + if (baseDir === nextBaseDir && sharedDocumentStore === nextSharedDocumentStore) return + baseDir = nextBaseDir + sharedDocumentStore = nextSharedDocumentStore userGlobalCache = null projectCache.clear() } export const isConfigured = (): boolean => baseDir !== null +export const isConfiguredFor = (dir: string): boolean => baseDir === path.resolve(dir) + // Reset to the unconfigured state (baseDir=null + caches cleared). `configure` is a process-global // setter with no other way back to null; tests that assert the UNCONFIGURED path (isConfigured()===false // → callers fall back to empty results) need this to guarantee their precondition regardless of a prior @@ -51,7 +57,8 @@ export const reset = (): void => { projectCache.clear() } -// Clear cached stores so a subsequent query re-reads from disk (after approve/reject/seed). +// Clear cached stores so a subsequent query re-reads from disk. Normal in-process writes must use +// the cached handles instead; this cold path is only for explicit external-change recovery/tests. export const invalidateCache = (): void => { userGlobalCache = null projectCache.clear() @@ -152,7 +159,10 @@ export const environmentFactAdoptionFor = (workspacePath: string): EnvironmentFa const base = ensureBase() const home = new DeepAgentCodeHome(base) const paths = home.ensureProject(projectIdForWorkspace(workspacePath), workspacePath) - return new EnvironmentFactAdoption(base, paths, workspacePath) + return new EnvironmentFactAdoption(base, paths, workspacePath, { + userGlobal: userGlobalStore(), + project: projectStore(workspacePath), + }) } // Open the project store for a workspace path. Throws if not configured. @@ -184,6 +194,10 @@ export const listAllForWorkspace = (workspacePath: string): readonly ReviewItem[ } return out } + +export const reviewSummaryForWorkspace = (workspacePath: string): { readonly pendingCount: number } => ({ + pendingCount: listByStatusForWorkspace(workspacePath, "candidate").filter((item) => item.type !== "skill").length, +}) export type ReviewItem = { readonly id: string readonly type: import("./document-store").DocType diff --git a/packages/core/test/deepagent/knowledge-source-cache.test.ts b/packages/core/test/deepagent/knowledge-source-cache.test.ts new file mode 100644 index 00000000..1a11f779 --- /dev/null +++ b/packages/core/test/deepagent/knowledge-source-cache.test.ts @@ -0,0 +1,67 @@ +import { afterEach, describe, expect, test } from "bun:test" +import { mkdtempSync, rmSync } from "node:fs" +import { tmpdir } from "node:os" +import path from "node:path" +import { + configure, + isConfiguredFor, + projectStoreFor, + reset, + reviewSummaryForWorkspace, + userGlobalStoreFor, +} from "../../src/deepagent/knowledge-source" + +const roots: string[] = [] + +const root = () => { + const value = mkdtempSync(path.join(tmpdir(), "deepagent-knowledge-cache-")) + roots.push(value) + return value +} + +afterEach(() => { + reset() + for (const value of roots.splice(0)) rmSync(value, { recursive: true, force: true }) +}) + +describe("knowledge source cache", () => { + test("reuses stores when the configured storage root is unchanged", () => { + const base = root() + configure(base) + const first = userGlobalStoreFor() + + configure(path.join(base, ".")) + + expect(isConfiguredFor(path.join(base, "."))).toBe(true) + expect(userGlobalStoreFor()).toBe(first) + }) + + test("replaces stores when the configured storage root changes", () => { + configure(root()) + const first = userGlobalStoreFor() + + configure(root()) + + expect(userGlobalStoreFor()).not.toBe(first) + }) + + test("summarizes pending review items from the live cache", () => { + const base = root() + const workspace = path.join(base, "workspace") + configure(base) + projectStoreFor(workspace).stageCandidate({ + type: "memory", + description: "review this learned behavior", + body: "review this learned behavior", + domain: "code", + scope: "project-shared", + projectId: "project-test", + sensitivity: "public", + risk: "low", + confidence: { evidence_strength: "medium", support_count: 1 }, + provenance: { source: "runner", run_ref: "run-1", evidence_refs: [] }, + }) + + expect(reviewSummaryForWorkspace(workspace)).toEqual({ pendingCount: 1 }) + }) +}) diff --git a/packages/deepagent-code/src/import/writer/memory.ts b/packages/deepagent-code/src/import/writer/memory.ts index b1760751..6dc14ffb 100644 --- a/packages/deepagent-code/src/import/writer/memory.ts +++ b/packages/deepagent-code/src/import/writer/memory.ts @@ -4,8 +4,9 @@ import { type DurableKnowledgeStore, openProjectStore, openUserGlobalStore, + projectIdForWorkspace, } from "@deepagent-code/core/deepagent/durable-knowledge-store" -import * as KnowledgeSource from "@deepagent-code/core/deepagent/knowledge-source" +import { isConfiguredFor, projectStoreFor, userGlobalStoreFor } from "@deepagent-code/core/deepagent/knowledge-source" import { classifyReview, DEFAULT_CONFIG } from "@deepagent-code/core/deepagent/auto-reviewer" import { looksSensitive } from "@deepagent-code/core/deepagent/memory-governance" import type { Doc } from "@deepagent-code/core/deepagent/document-store" @@ -35,7 +36,13 @@ export function stageAndReviewMemories(memories: MemoryItem[], baseDir: string): let staged = 0 for (const item of memories) { - const store = item.cwd ? openProjectStore(baseDir, item.cwd) : openUserGlobalStore(baseDir) + const store = isConfiguredFor(baseDir) + ? item.cwd + ? projectStoreFor(item.cwd) + : userGlobalStoreFor() + : item.cwd + ? openProjectStore(baseDir, item.cwd) + : openUserGlobalStore(baseDir) // Dedup store instances so the review pass walks each store once even when // many memories share the same root (cwd or user-global). const storeKey = item.cwd ?? "__global__" @@ -49,6 +56,7 @@ export function stageAndReviewMemories(memories: MemoryItem[], baseDir: string): body: item.body, domain: null, scope: item.cwd ? "project-shared" : "user-global", + ...(item.cwd ? { projectId: projectIdForWorkspace(item.cwd) } : {}), sensitivity: "public", risk: "low", confidence: { evidence_strength: "weak", support_count: 1 }, @@ -60,7 +68,6 @@ export function stageAndReviewMemories(memories: MemoryItem[], baseDir: string): } const { approved, pending } = autoReviewMemories(stores) - invalidateKnowledgeCache() return { staged, writtenToInstructions: false, approved, pending } } @@ -120,14 +127,6 @@ function shouldAutoApprove(doc: Doc): boolean { return true } -function invalidateKnowledgeCache(): void { - try { - KnowledgeSource.invalidateCache() - } catch { - /* knowledge-source not configured in this process (e.g. CLI) — safe to skip */ - } -} - /** * Fallback / always-on path: append imported memories to an AGENTS.md the * instruction-context loader auto-reads, so they are immediately visible as diff --git a/packages/deepagent-code/src/server/routes/instance/httpapi/groups/deepagent.ts b/packages/deepagent-code/src/server/routes/instance/httpapi/groups/deepagent.ts index 905666fe..b074359e 100644 --- a/packages/deepagent-code/src/server/routes/instance/httpapi/groups/deepagent.ts +++ b/packages/deepagent-code/src/server/routes/instance/httpapi/groups/deepagent.ts @@ -136,6 +136,8 @@ export const DeepAgentKnowledgeItem = Schema.Struct({ export const DeepAgentKnowledgeList = Schema.Struct({ items: Schema.Array(DeepAgentKnowledgeItem) }) +export const DeepAgentKnowledgeReviewSummary = Schema.Struct({ pendingCount: Schema.Number }) + export const DeepAgentKnowledgeStatusInput = Schema.Struct({ ids: Schema.Array(Schema.String) }) export const DeepAgentKnowledgeStatusResult = Schema.Struct({ updated: Schema.Array(Schema.String) }) @@ -507,6 +509,19 @@ export const DeepAgentApi = HttpApi.make("deepagent").add( }), ), ) + .add( + HttpApiEndpoint.get("knowledgeReviewSummary", `${root}/knowledge/review-summary`, { + query: WorkspaceRoutingQuery, + success: described(DeepAgentKnowledgeReviewSummary, "Pending durable knowledge count for the Review badge"), + error: DeepAgentPromotionError, + }).annotateMerge( + OpenApi.annotations({ + identifier: "deepagent.knowledge.reviewSummary", + summary: "Get the DeepAgent knowledge review summary", + description: "Return the pending review count without projecting the complete knowledge review list.", + }), + ), + ) .add( HttpApiEndpoint.post("knowledgeApprove", `${root}/knowledge/approve`, { query: WorkspaceRoutingQuery, @@ -656,7 +671,10 @@ export const DeepAgentApi = HttpApi.make("deepagent").add( .add( HttpApiEndpoint.get("panelStatus", `${root}/panel/status`, { query: Schema.Struct({ ...WorkspaceRoutingQueryFields, sessionID: Schema.String }), - success: described(DeepAgentPanelStatusResult, "Effective panel armed state (explicit toggle or global default)"), + success: described( + DeepAgentPanelStatusResult, + "Effective panel armed state (explicit toggle or global default)", + ), error: DeepAgentPromotionError, }).annotateMerge( OpenApi.annotations({ diff --git a/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/deepagent.ts b/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/deepagent.ts index 210d9e9b..caebfe05 100644 --- a/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/deepagent.ts +++ b/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/deepagent.ts @@ -60,9 +60,11 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen model: { providerID: model.providerID, modelID: model.modelID }, }) return (turnInput) => - runTurn({ agentType: turnInput.agentType, prompt: turnInput.prompt, outputSchema: turnInput.outputSchema }).pipe( - Effect.map((r) => ({ structured: r.structured })), - ) + runTurn({ + agentType: turnInput.agentType, + prompt: turnInput.prompt, + outputSchema: turnInput.outputSchema, + }).pipe(Effect.map((r) => ({ structured: r.structured }))) }) const resolveReviewRunsDir = Effect.fn("DeepAgentHttpApi.resolveReviewRunsDir")(function* () { @@ -147,7 +149,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen record, AgentGateway.DeepAgentKnowledgeSource.userGlobalStoreFor(), ) - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return record }, catch: (error) => @@ -172,7 +173,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen ctx.payload.candidate.candidate_id, "rejected", ) - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return { candidateId: ctx.payload.candidate.candidate_id, fingerprint, reason: ctx.payload.reason } }, catch: (error) => @@ -213,13 +213,21 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen }) }) + const knowledgeReviewSummary = Effect.fn("DeepAgentHttpApi.knowledgeReviewSummary")(function* () { + const dir = yield* workspaceDir() + return yield* Effect.try({ + try: () => AgentGateway.DeepAgentKnowledgeSource.reviewSummaryForWorkspace(dir), + catch: (error) => + new DeepAgentPromotionError({ message: error instanceof Error ? error.message : String(error) }), + }) + }) + const knowledgeApprove = Effect.fn("DeepAgentHttpApi.knowledgeApprove")(function* (ctx) { const dir = yield* workspaceDir() return yield* Effect.try({ try: () => { for (const id of ctx.payload.ids) AgentGateway.DeepAgentKnowledgeSource.setApprovalForWorkspace(dir, id, "approved") - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return { updated: ctx.payload.ids } }, catch: (error) => @@ -233,7 +241,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen try: () => { for (const id of ctx.payload.ids) AgentGateway.DeepAgentKnowledgeSource.setApprovalForWorkspace(dir, id, "rejected") - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return { updated: ctx.payload.ids } }, catch: (error) => @@ -273,7 +280,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen if (AgentGateway.DeepAgentKnowledgeSource.setApprovalForWorkspace(dir, ref, "rejected")) demoted.push(ref) else notInStore.push(ref) } - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() } return { ship: decision.ship, @@ -415,8 +421,8 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen }) // V3.8.1 §G environment-fact use-gate handlers. The adoption service roots at the same gateway - // baseDir the retriever reads (workspaceDir() calls configureGateway first), keyed by the active - // workspace path — so a project's adopt/reject decisions are isolated per project (§G.8). + // baseDir the retriever reads, keyed by the active workspace path — so a project's adopt/reject + // decisions are isolated per project (§G.8). const now = () => new Date().toISOString() const envFacts = Effect.fn("DeepAgentHttpApi.envFacts")(function* () { @@ -435,7 +441,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen const adoption = AgentGateway.DeepAgentKnowledgeSource.environmentFactAdoptionFor(dir) if (ctx.payload.decision === "adopt") adoption.adopt(ctx.payload.factId, now()) else adoption.reject(ctx.payload.factId, now()) - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return { ok: true, factId: ctx.payload.factId } }, catch: (error) => @@ -456,7 +461,6 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen mode: ctx.payload.mode, now: now(), }) - AgentGateway.DeepAgentKnowledgeRetriever.invalidateCache() return { ok: true, factId: updatedId } }, catch: (error) => @@ -504,7 +508,8 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen const verdict = yield* consultPanel( { question: - ctx.payload.question ?? "Review the current changes in this conversation for correctness, security, and design.", + ctx.payload.question ?? + "Review the current changes in this conversation for correctness, security, and design.", codeRefs: ctx.payload.codeRefs ? [...ctx.payload.codeRefs] : [], parentSessionID: sessionID, ...(ctx.payload.lenses ? { lenses: [...ctx.payload.lenses] } : {}), @@ -519,9 +524,7 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen // The global Expert Panel default (§C): the effective armed state falls back to this when a session // has never explicitly toggled. Read from the first-party SettingsStore (expertPanelDefault). const expertPanelDefault = () => - Effect.promise(() => SettingsStore.read()).pipe( - Effect.map((s) => s.deepagent?.expertPanelDefault ?? false), - ) + Effect.promise(() => SettingsStore.read()).pipe(Effect.map((s) => s.deepagent?.expertPanelDefault ?? false)) const panelArm = Effect.fn("DeepAgentHttpApi.panelArm")(function* (ctx) { const { sessionID, armed, rounds } = ctx.payload @@ -701,7 +704,9 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen ...(ctx.query.scope ? { scope: ctx.query.scope } : {}), }) index.close() - return { hits: hits.map((h) => ({ docId: h.docId, type: h.type, scope: h.scope, title: h.title, score: h.score })) } + return { + hits: hits.map((h) => ({ docId: h.docId, type: h.type, scope: h.scope, title: h.title, score: h.score })), + } }) const wikiEdit = Effect.fn("DeepAgentHttpApi.wikiEdit")(function* (ctx) { @@ -748,6 +753,7 @@ export const deepagentHandlers = HttpApiBuilder.group(InstanceHttpApi, "deepagen .handle("promote", promote) .handle("reject", reject) .handle("knowledgePending", knowledgePending) + .handle("knowledgeReviewSummary", knowledgeReviewSummary) .handle("knowledgeApprove", knowledgeApprove) .handle("knowledgeRejectIds", knowledgeRejectIds) .handle("knowledgeShipGate", knowledgeShipGate) diff --git a/packages/deepagent-code/test/deepagent/knowledge-import-cache.test.ts b/packages/deepagent-code/test/deepagent/knowledge-import-cache.test.ts new file mode 100644 index 00000000..0578dd73 --- /dev/null +++ b/packages/deepagent-code/test/deepagent/knowledge-import-cache.test.ts @@ -0,0 +1,65 @@ +import { afterEach, describe, expect, test } from "bun:test" +import { mkdtempSync, rmSync } from "node:fs" +import { tmpdir } from "node:os" +import path from "node:path" +import { + configure, + reset, + reviewSummaryForWorkspace, + userGlobalStoreFor, +} from "@deepagent-code/core/deepagent/knowledge-source" +import { stageAndReviewMemories } from "../../src/import/writer/memory" + +const roots: string[] = [] + +afterEach(() => { + reset() + for (const value of roots.splice(0)) rmSync(value, { recursive: true, force: true }) +}) + +describe("knowledge import cache", () => { + test("writes through the configured store without invalidating it", () => { + const base = mkdtempSync(path.join(tmpdir(), "deepagent-knowledge-import-cache-")) + roots.push(base) + configure(base) + const store = userGlobalStoreFor() + + expect( + stageAndReviewMemories( + [ + { + source: "codex", + slug: "safe-memory", + title: "Safe memory", + body: "Use the repository typecheck command before submitting a change.", + }, + ], + base, + ), + ).toEqual({ staged: 1, writtenToInstructions: false, approved: 1, pending: 0 }) + expect(userGlobalStoreFor()).toBe(store) + }) + + test("makes imported project candidates visible in the live review summary", () => { + const base = mkdtempSync(path.join(tmpdir(), "deepagent-knowledge-import-cache-")) + const workspace = path.join(base, "workspace") + roots.push(base) + configure(base) + + const result = stageAndReviewMemories( + [ + { + source: "codex", + slug: "secret-memory", + title: "Secret memory", + body: "API_TOKEN=example-secret-value-that-needs-human-review", + cwd: workspace, + }, + ], + base, + ) + + expect(result.pending).toBe(1) + expect(reviewSummaryForWorkspace(workspace)).toEqual({ pendingCount: 1 }) + }) +}) From 551b6cb4c7c156ae26c6dd6aff1ede55601db239 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:01:32 +0800 Subject: [PATCH 13/22] feat(app): group wiki content by graph --- .../src/components/wiki/dialog-wiki.test.ts | 22 +++ .../app/src/components/wiki/dialog-wiki.tsx | 139 ++++++++++++------ 2 files changed, 120 insertions(+), 41 deletions(-) create mode 100644 packages/app/src/components/wiki/dialog-wiki.test.ts diff --git a/packages/app/src/components/wiki/dialog-wiki.test.ts b/packages/app/src/components/wiki/dialog-wiki.test.ts new file mode 100644 index 00000000..7ed49225 --- /dev/null +++ b/packages/app/src/components/wiki/dialog-wiki.test.ts @@ -0,0 +1,22 @@ +import { describe, expect, test } from "bun:test" +import { resolveWikiExpandedGroup, type WikiGroup } from "./dialog-wiki" + +const groups = (counts: Partial>) => + (["knowledge", "memory", "code", "document"] as const).map((group) => ({ + group, + items: Array.from({ length: counts[group] ?? 0 }, (_, index) => ({ id: index })), + })) + +describe("wiki graph expansion", () => { + test("search reveals the first graph containing a result", () => { + expect(resolveWikiExpandedGroup(groups({ code: 2, document: 1 }), "knowledge", true)).toBe("code") + }) + + test("empty search results close the prior graph instead of showing a misleading empty group", () => { + expect(resolveWikiExpandedGroup(groups({}), "knowledge", true)).toBeUndefined() + }) + + test("manual expansion is preserved outside search", () => { + expect(resolveWikiExpandedGroup(groups({ document: 1 }), "memory", false)).toBe("memory") + }) +}) diff --git a/packages/app/src/components/wiki/dialog-wiki.tsx b/packages/app/src/components/wiki/dialog-wiki.tsx index 3d9170b8..f2ec1e87 100644 --- a/packages/app/src/components/wiki/dialog-wiki.tsx +++ b/packages/app/src/components/wiki/dialog-wiki.tsx @@ -1,6 +1,7 @@ -import { Component, createMemo, createResource, createSignal, For, Show } from "solid-js" +import { Component, createEffect, createMemo, createResource, createSignal, For, Show } from "solid-js" import { Dialog } from "@deepagent-code/ui/v2/dialog-v2" import { Button } from "@deepagent-code/ui/button" +import { Collapsible } from "@deepagent-code/ui/collapsible" import { Icon } from "@deepagent-code/ui/icon" import { Markdown } from "@deepagent-code/ui/markdown" import { useLanguage } from "@/context/language" @@ -28,22 +29,30 @@ export { } from "./wiki.api" // §B.2 governance grouping: two governable (Knowledge/Memory), two monitor-only (Document/Code). -const TYPE_GROUP: Record = { +export type WikiGroup = "knowledge" | "memory" | "code" | "document" + +const TYPE_GROUP: Record = { knowledge: "knowledge", strategy: "knowledge", methodology: "knowledge", memory: "memory", code_symbol: "code", } -const groupOf = (type: string): "knowledge" | "memory" | "code" | "document" => - (TYPE_GROUP[type] as "knowledge" | "memory" | "code") ?? "document" +const groupOf = (type: string): WikiGroup => TYPE_GROUP[type] ?? "document" + +const GROUP_ORDER: readonly WikiGroup[] = ["knowledge", "memory", "code", "document"] +const GROUP_ICON = { + knowledge: "knowledge-check", + memory: "brain", + code: "code-lines", + document: "open-file", +} as const -const GROUP_ORDER: ReadonlyArray<"knowledge" | "memory" | "document" | "code"> = [ - "knowledge", - "memory", - "document", - "code", -] +export const resolveWikiExpandedGroup = ( + groups: readonly { group: WikiGroup; items: readonly unknown[] }[], + expanded: WikiGroup | undefined, + searching: boolean, +) => (searching ? groups.find((group) => group.items.length > 0)?.group : expanded) export const DialogWiki: Component<{ client: WikiClient }> = (props) => { const language = useLanguage() @@ -83,7 +92,7 @@ export const DialogWiki: Component<{ client: WikiClient }> = (props) => { list.push(p) byGroup.set(g, list) } - return GROUP_ORDER.map((g) => ({ group: g, items: byGroup.get(g) ?? [] })).filter((x) => x.items.length > 0) + return GROUP_ORDER.map((group) => ({ group, items: byGroup.get(group) ?? [] })) }) // The rendered detail page for the current selection. @@ -157,10 +166,11 @@ export const DialogWiki: Component<{ client: WikiClient }> = (props) => { language.t(`wiki.type.${groupOf(t)}`)} + typeLabel={(group) => language.t(`wiki.type.${group}`)} /> = (props) => { } const WikiList: Component<{ - groups: { group: string; items: WikiPageSummary[] }[] + groups: { group: WikiGroup; items: WikiPageSummary[] }[] loading: boolean + searchKey: string empty: string selectedId: string | undefined onSelect: (p: WikiPageSummary) => void - typeLabel: (type: string) => string + typeLabel: (group: WikiGroup) => string }> = (props) => { const language = useLanguage() + const [expanded, setExpanded] = createSignal() + createEffect(() => { + const searchKey = props.searchKey + const groups = props.groups + if (!searchKey) return + setExpanded(resolveWikiExpandedGroup(groups, undefined, true)) + }) return ( -
+
{language.t("wiki.loading")}
} > - 0} - fallback={
{props.empty}
} - > +
{(group) => ( -
- - {props.typeLabel(group.items[0]!.type)} - - - {(p) => ( - - )} - -
+ + {(p) => ( + + )} + + +
+ + )} - +
) @@ -262,7 +314,12 @@ const WikiDetail: Component<{ {language.t("wiki.version", { version: pg().version })} - From d0060da093e70ee019f0773d0d0796eaae34f878 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:01:57 +0800 Subject: [PATCH 14/22] fix(app): preserve project sidebar identity --- .../app/src/pages/layout/sidebar-project.tsx | 28 +++++++-------- .../app/src/pages/layout/sidebar-shell.tsx | 36 ++++--------------- 2 files changed, 20 insertions(+), 44 deletions(-) diff --git a/packages/app/src/pages/layout/sidebar-project.tsx b/packages/app/src/pages/layout/sidebar-project.tsx index 0337e02f..5e603603 100644 --- a/packages/app/src/pages/layout/sidebar-project.tsx +++ b/packages/app/src/pages/layout/sidebar-project.tsx @@ -269,24 +269,24 @@ const ProjectPreviewPanel = (props: { ) export const SortableProject = (props: { - project: LocalProject + project: Accessor mobile?: boolean ctx: ProjectSidebarContext sortNow: Accessor }): JSX.Element => { const serverSync = useServerSync() const language = useLanguage() - const sortable = createSortable(props.project.worktree) - const selected = createMemo(() => props.ctx.currentProject()?.worktree === props.project.worktree) - const workspaces = createMemo(() => props.ctx.workspaceIds(props.project).slice(0, 2)) - const workspaceEnabled = createMemo(() => props.ctx.workspacesEnabled(props.project)) - const dirs = createMemo(() => props.ctx.workspaceIds(props.project)) + const sortable = createSortable(props.project().worktree) + const selected = createMemo(() => props.ctx.currentProject()?.worktree === props.project().worktree) + const workspaces = createMemo(() => props.ctx.workspaceIds(props.project()).slice(0, 2)) + const workspaceEnabled = createMemo(() => props.ctx.workspacesEnabled(props.project())) + const dirs = createMemo(() => props.ctx.workspaceIds(props.project())) const [state, setState] = createStore({ menu: false, suppressHover: false, }) - const isHoverProject = () => props.ctx.hoverProject() === props.project.worktree + const isHoverProject = () => props.ctx.hoverProject() === props.project().worktree const preview = createMemo(() => !props.mobile && props.ctx.sidebarOpened()) const overlay = createMemo(() => !props.mobile && !props.ctx.sidebarOpened()) const active = createMemo(() => state.menu || (preview() ? isHoverProject() : overlay() && isHoverProject())) @@ -296,12 +296,12 @@ export const SortableProject = (props: { const label = (directory: string) => { const [data] = serverSync.child(directory, { bootstrap: false }) const kind = - directory === props.project.worktree ? language.t("workspace.type.local") : language.t("workspace.type.sandbox") - const name = props.ctx.workspaceLabel(directory, data.vcs?.branch, props.project.id) + directory === props.project().worktree ? language.t("workspace.type.local") : language.t("workspace.type.sandbox") + const name = props.ctx.workspaceLabel(directory, data.vcs?.branch, props.project().id) return `${kind} : ${name}` } - const projectStore = createMemo(() => serverSync.child(props.project.worktree, { bootstrap: false })[0]) + const projectStore = createMemo(() => serverSync.child(props.project().worktree, { bootstrap: false })[0]) const isWorking = createMemo(() => dirs().some((directory) => { const [store] = serverSync.child(directory, { bootstrap: false }) @@ -315,7 +315,7 @@ export const SortableProject = (props: { } const tile = () => ( setState("menu", value)} - setOpen={(value) => props.ctx.onHoverOpenChanged(props.project.worktree, value)} + setOpen={(value) => props.ctx.onHoverOpenChanged(props.project().worktree, value)} setSuppressHover={(value) => setState("suppressHover", value)} language={language} /> @@ -353,11 +353,11 @@ export const SortableProject = (props: { onOpenChange={(value) => { if (state.menu) return if (value && state.suppressHover) return - props.ctx.onHoverOpenChanged(props.project.worktree, value) + props.ctx.onHoverOpenChanged(props.project().worktree, value) }} > aimMove: (event: MouseEvent) => void projects: Accessor - renderProject: (project: LocalProject) => JSX.Element + renderProject: (project: Accessor) => JSX.Element handleDragStart: (event: unknown) => void handleDragEnd: () => void handleDragOver: (event: DragEvent) => void @@ -75,30 +69,12 @@ export const SidebarContent = (props: {
p.worktree)}> - {/* - * Iterate over stable worktree STRINGS, not project objects. - * - * layout.projects.list() creates new object references on every - * memo run (via .map({ ...project, icon })). Using those objects - * as For keys causes SolidJS to destroy+recreate EVERY SortableProject - * on each icon-color update or navigation, which: - * 1. Mass-fires DnD cleanup before DragDropProvider is gone → errors - * 2. Triggers every SortableWorkspace's bootstrap API call at once → 3s lag - * - * Worktree strings are stable primitives (same path === same identity). - * The project data updates reactively inside each component via the - * `projectFor` accessor below. - */} + {/* Keep component identity on worktree; project objects are rebuilt by list(). */} p.worktree)}> {(worktree) => { - const project = createMemo( - () => props.projects().find((p) => p.worktree === worktree), - ) - return ( - - {(p) => props.renderProject(p)} - - ) + const initial = props.projects().find((p) => p.worktree === worktree)! + const project = createMemo(() => props.projects().find((p) => p.worktree === worktree) ?? initial) + return props.renderProject(project) }} From 43a3f4ebb87281c1afcad3b3cdcebdab9caf7d60 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:02:56 +0800 Subject: [PATCH 15/22] feat(llm): support safe freeform patch tools --- packages/core/src/tool/tool.ts | 2 +- packages/core/test/tool-bash.test.ts | 5 +- packages/core/test/tool-edit.test.ts | 5 +- packages/core/test/tool-write.test.ts | 5 +- packages/deepagent-code/src/session/llm.ts | 188 ++++++++++++-- .../deepagent-code/src/session/llm/ai-sdk.ts | 17 +- .../src/session/llm/freeform-tools.ts | 41 +++ .../src/session/llm/native-runtime.ts | 14 +- .../src/tool/apply-patch-grammar.ts | 20 ++ .../deepagent-code/src/tool/apply_patch.txt | 1 + .../src/tool/apply_patch_chunk.ts | 146 +++++++++++ packages/deepagent-code/src/tool/edit.txt | 1 + packages/deepagent-code/src/tool/registry.ts | 9 +- packages/deepagent-code/src/tool/write.txt | 1 + .../test/session/llm-native.test.ts | 78 +++++- .../deepagent-code/test/session/llm.test.ts | 130 +++++++++- .../test/tool/apply_patch_chunk.test.ts | 131 ++++++++++ packages/llm/src/cache-policy.ts | 2 +- .../llm/src/protocols/anthropic-messages.ts | 47 ++-- .../llm/src/protocols/bedrock-converse.ts | 12 +- packages/llm/src/protocols/gemini.ts | 7 +- packages/llm/src/protocols/openai-chat.ts | 11 +- .../llm/src/protocols/openai-responses.ts | 206 +++++++++++---- .../llm/src/protocols/utils/tool-stream.ts | 6 +- packages/llm/src/schema/messages.ts | 55 +++- packages/llm/src/tool.ts | 47 ++-- packages/llm/test/generate-object.test.ts | 3 +- packages/llm/test/llm.test.ts | 13 +- .../test/provider/anthropic-messages.test.ts | 38 ++- .../llm/test/provider/openai-chat.test.ts | 41 ++- .../test/provider/openai-responses.test.ts | 241 +++++++++++++++++- 31 files changed, 1386 insertions(+), 137 deletions(-) create mode 100644 packages/deepagent-code/src/session/llm/freeform-tools.ts create mode 100644 packages/deepagent-code/src/tool/apply-patch-grammar.ts create mode 100644 packages/deepagent-code/src/tool/apply_patch_chunk.ts create mode 100644 packages/deepagent-code/test/tool/apply_patch_chunk.test.ts diff --git a/packages/core/src/tool/tool.ts b/packages/core/src/tool/tool.ts index 01834713..1372d71f 100644 --- a/packages/core/src/tool/tool.ts +++ b/packages/core/src/tool/tool.ts @@ -68,7 +68,7 @@ export function make, Output extends SchemaType { const cached = definitions.get(name) if (cached) return cached - const definition = new ToolDefinition({ + const definition = ToolDefinition.make({ name, description: config.description, inputSchema: toJsonSchema(config.input), diff --git a/packages/core/test/tool-bash.test.ts b/packages/core/test/tool-bash.test.ts index 5e6b0e7f..1b09f72b 100644 --- a/packages/core/test/tool-bash.test.ts +++ b/packages/core/test/tool-bash.test.ts @@ -156,7 +156,10 @@ describe("BashTool", () => { Effect.gen(function* () { const definitions = yield* toolDefinitions(registry) expect(definitions.map((tool) => tool.name)).toEqual(["bash"]) - expect(definitions[0]?.inputSchema).not.toHaveProperty("properties.background") + const definition = definitions[0] + expect(definition?.type).toBe("function") + if (!definition || definition.type !== "function") throw new Error("bash must register as a function tool") + expect(definition.inputSchema).not.toHaveProperty("properties.background") expect(yield* toolDefinitions(registry, [{ action: "bash", resource: "*", effect: "deny" }])).toEqual([]) expect( yield* settleTool(registry, call({ command: "pwd", description: "Print working directory" })), diff --git a/packages/core/test/tool-edit.test.ts b/packages/core/test/tool-edit.test.ts index cd9ea679..8604cc12 100644 --- a/packages/core/test/tool-edit.test.ts +++ b/packages/core/test/tool-edit.test.ts @@ -400,7 +400,10 @@ test("keeps the locked edit schema, semantics docstring, and deferred TODOs visi const definition = await Effect.runPromise( withTool(path.dirname(fileURLToPath(import.meta.url)), (registry) => toolDefinitions(registry)), ) - const schema = definition[0]?.inputSchema as { readonly properties?: Record } + const tool = definition[0] + expect(tool?.type).toBe("function") + if (!tool || tool.type !== "function") throw new Error("edit must register as a function tool") + const schema = tool.inputSchema as { readonly properties?: Record } expect(Object.keys(schema.properties ?? {}).sort()).toEqual(["newString", "oldString", "path", "replaceAll"]) expect(source).toContain( diff --git a/packages/core/test/tool-write.test.ts b/packages/core/test/tool-write.test.ts index e7da617f..54ffa416 100644 --- a/packages/core/test/tool-write.test.ts +++ b/packages/core/test/tool-write.test.ts @@ -277,7 +277,10 @@ test("keeps the locked write schema, semantics docstring, and deferred UX TODOs const definition = await Effect.runPromise( withTool(path.dirname(fileURLToPath(import.meta.url)), (registry) => toolDefinitions(registry)), ) - const schema = definition[0]?.inputSchema as { readonly properties?: Record } + const tool = definition[0] + expect(tool?.type).toBe("function") + if (!tool || tool.type !== "function") throw new Error("write must register as a function tool") + const schema = tool.inputSchema as { readonly properties?: Record } expect(Object.keys(schema.properties ?? {}).sort()).toEqual(["content", "path"]) expect(source).toContain( diff --git a/packages/deepagent-code/src/session/llm.ts b/packages/deepagent-code/src/session/llm.ts index 87d9f0ac..13b29c23 100644 --- a/packages/deepagent-code/src/session/llm.ts +++ b/packages/deepagent-code/src/session/llm.ts @@ -6,7 +6,15 @@ import { Log } from "@deepagent-code/core/util/log" import { Global } from "@deepagent-code/core/global" import { Context, Effect, Layer } from "effect" import * as Stream from "effect/Stream" -import { streamText, wrapLanguageModel, type ModelMessage, type Tool, APICallError, NoSuchToolError, InvalidToolInputError } from "ai" +import { + streamText, + wrapLanguageModel, + type ModelMessage, + type Tool, + APICallError, + NoSuchToolError, + InvalidToolInputError, +} from "ai" import { type LLMEvent } from "@deepagent-code/llm" import { AgentGateway } from "@deepagent-code/core/agent-gateway" import { LLMClient, RequestExecutor, WebSocketExecutor } from "@deepagent-code/llm/route" @@ -30,6 +38,7 @@ import * as OtelTracer from "@effect/opentelemetry/Tracer" import { LLMAISDK } from "./llm/ai-sdk" import { LLMNativeRuntime } from "./llm/native-runtime" import { LLMRequestPrep } from "./llm/request" +import { FreeformTools } from "./llm/freeform-tools" import { configureGateway } from "@/deepagent/config" const log = Log.create({ service: "llm" }) @@ -41,23 +50,137 @@ const deepagentModelAuthProviderID = (model: Provider.Model) => { return typeof value === "string" && value.length > 0 ? value : undefined } -// Detect whether the resolved request options enable extended thinking / reasoning. -// Providers reject `tool_choice: required/object` while thinking is active (e.g. -// "The tool_choice parameter does not support being set to required or object in -// thinking mode"), which broke structured-output subagent calls. When thinking is -// on, the caller downgrades toolChoice to auto and relies on the schema-aware -// system prompt (buildStructuredOutputSystemPrompt) to elicit the structured tool -// call instead of forcing it. -const thinkingActive = (options: Record | undefined): boolean => { - if (!options || typeof options !== "object") return false - const effort = options.reasoningEffort ?? options.reasoning?.effort +// Decide from the fully merged per-request options. Some providers reject +// `tool_choice: required/object` while thinking is active. Callers that need a +// hard guarantee must either disable thinking for this turn or use an explicitly +// bounded auto-only controller; this layer never silently weakens `required`. +const isRecord = (value: unknown): value is Record => + typeof value === "object" && value !== null && !Array.isArray(value) + +const thinkingActive = (options: Record | undefined): boolean => { + if (!options) return false + const effort = options.reasoningEffort ?? (isRecord(options.reasoning) ? options.reasoning.effort : undefined) if (typeof effort === "string" && effort !== "none") return true const thinking = options.thinking - if (thinking && typeof thinking === "object" && thinking.type !== "disabled") return true - if (options.thinkingConfig && typeof options.thinkingConfig === "object") return true + if (isRecord(thinking) && thinking.type !== "disabled") return true + if (isRecord(options.thinkingConfig)) return true return false } +export type ToolChoiceProtocol = + | "openai_responses" + | "openai_chat" + | "anthropic_messages" + | "gemini" + | "bedrock_converse" + | "unknown" + +const OPENAI_RESPONSES_PACKAGES = new Set([ + "@ai-sdk/openai", + "@ai-sdk/azure", + "@ai-sdk/amazon-bedrock/mantle", + "@ai-sdk/xai", +]) +const OPENAI_CHAT_PACKAGES = new Set([ + "@ai-sdk/openai-compatible", + "@openrouter/ai-sdk-provider", + "@ai-sdk/groq", + "@ai-sdk/deepinfra", + "@ai-sdk/cerebras", + "@ai-sdk/togetherai", + "@ai-sdk/perplexity", + "@ai-sdk/vercel", + "@ai-sdk/alibaba", + "@ai-sdk/github-copilot", + "venice-ai-sdk-provider", +]) + +export function toolChoiceProtocol(model: Provider.Model): ToolChoiceProtocol { + if (OPENAI_RESPONSES_PACKAGES.has(model.api.npm)) return "openai_responses" + if (OPENAI_CHAT_PACKAGES.has(model.api.npm)) return "openai_chat" + if (model.api.npm === "@ai-sdk/anthropic" || model.api.npm === "@ai-sdk/google-vertex/anthropic") + return "anthropic_messages" + if (model.api.npm === "@ai-sdk/google" || model.api.npm === "@ai-sdk/google-vertex") return "gemini" + if (model.api.npm === "@ai-sdk/amazon-bedrock") return "bedrock_converse" + return "unknown" +} + +function reasoningOnly(model: Provider.Model) { + if (model.options?.reasoningOnly === true) return true + if (!model.capabilities.reasoning) return false + const id = `${model.id} ${model.api.id}`.toLowerCase() + return ( + /\bo(?:1|3|4)(?:\b|[._-])/.test(id) || + id.includes("gpt-5-pro") || + id.includes("deepseek-r1") || + id.includes("deepseek-reasoner") + ) +} + +export function finalizerCapability(model: Provider.Model) { + const protocol = toolChoiceProtocol(model) + if (!model.capabilities.toolcall) + return { + capability: "unsupported" as const, + protocol, + reasoning: "inherit" as const, + reason: "model_has_no_tool_call_capability" as const, + } + if (reasoningOnly(model)) + return { + capability: "auto_only" as const, + protocol, + reasoning: "inherit" as const, + toolChoice: "auto" as const, + reason: "reasoning_cannot_be_disabled" as const, + } + if (protocol === "unknown") + return { + capability: "auto_only" as const, + protocol, + reasoning: "inherit" as const, + toolChoice: "auto" as const, + reason: "protocol_forced_tool_unverified" as const, + } + return { + capability: "forced_tool" as const, + protocol, + reasoning: "disabled" as const, + toolChoice: "required" as const, + } +} + +export const disableThinking = (options: Record | undefined): Record | undefined => { + if (!options) return options + const result = { ...options } + delete result.reasoningEffort + delete result.reasoning + delete result.thinking + delete result.thinkingConfig + if ("enable_thinking" in result) result.enable_thinking = false + if (isRecord(result.chat_template_args)) { + result.chat_template_args = { ...result.chat_template_args, enable_thinking: false } + } + if (isRecord(result.modelParams)) { + result.modelParams = disableThinking(result.modelParams) + } + return result +} + +export function decideToolChoice( + toolChoice: StreamInput["toolChoice"], + options: Record | undefined, + model?: Provider.Model, +): + | { capability: "forced_tool" | "auto_only"; toolChoice: StreamInput["toolChoice"] } + | { capability: "unsupported_thinking_with_forced_tool" } + | { capability: "unsupported_forced_tool" } { + if (toolChoice !== "required") return { capability: "auto_only", toolChoice } + if (model && toolChoiceProtocol(model) === "unknown") return { capability: "unsupported_forced_tool" } + if (!thinkingActive(options)) return { capability: "forced_tool", toolChoice } + return { capability: "unsupported_thinking_with_forced_tool" } +} + export type StreamInput = { user: SessionV1.User sessionID: string @@ -71,6 +194,7 @@ export type StreamInput = { tools: Record retries?: number toolChoice?: "auto" | "required" | "none" + reasoning?: "inherit" | "disabled" } export type StreamRequest = StreamInput & { @@ -153,8 +277,10 @@ const live: Layer.Layer< const providerMaxRetries = typeof item.options?.maxRetries === "number" ? item.options.maxRetries : undefined const isWorkflow = language instanceof GitLabWorkflowLanguageModel + const protocolTools = isWorkflow ? input.tools : FreeformTools.tools(language, input.tools) const prepared = yield* LLMRequestPrep.prepare({ ...input, + tools: protocolTools, provider: item, auth: info, plugin, @@ -169,10 +295,24 @@ const live: Layer.Layer< }, }) - // Structured-output callers force toolChoice:"required", which providers reject - // while thinking is active. Downgrade to auto in that case so the call still - // succeeds; the schema-aware system prompt keeps the model on the structured path. - const effectiveToolChoice = thinkingActive(prepared.params.options) ? undefined : input.toolChoice + const effectiveOptions = + input.reasoning === "disabled" ? disableThinking(prepared.params.options) : prepared.params.options + const toolChoiceDecision = decideToolChoice(input.toolChoice, effectiveOptions, input.model) + if (toolChoiceDecision.capability === "unsupported_forced_tool") { + return yield* Effect.fail( + new Error( + `[unsupported_forced_tool] Required tool choice is not verified for provider protocol ${toolChoiceProtocol(input.model)}. Use an explicitly bounded auto-only controller.`, + ), + ) + } + if (toolChoiceDecision.capability === "unsupported_thinking_with_forced_tool") { + return yield* Effect.fail( + new Error( + "[unsupported_thinking_with_forced_tool] Required tool choice cannot be used while reasoning is active. Disable reasoning for this turn or use an explicitly bounded auto-only controller.", + ), + ) + } + const effectiveToolChoice = toolChoiceDecision.toolChoice // Wire up toolExecutor for DWS workflow models so that tool calls // from the workflow service are executed via deepagent-code's tool system @@ -308,6 +448,8 @@ const live: Layer.Layer< }) } + const runtimeTools = prepared.tools + const tracer = cfg.experimental?.openTelemetry ? Option.getOrUndefined(yield* Effect.serviceOption(OtelTracer.OtelTracer)) : undefined @@ -333,13 +475,13 @@ const live: Layer.Layer< auth: info, llmClient, messages: prepared.messages, - tools: prepared.tools, + tools: runtimeTools, toolChoice: effectiveToolChoice, temperature: prepared.params.temperature, topP: prepared.params.topP, topK: prepared.params.topK, maxOutputTokens: prepared.params.maxOutputTokens, - providerOptions: prepared.params.options, + providerOptions: effectiveOptions, headers: prepared.headers, abort: input.abort, metadata: prepared.metadata, @@ -408,7 +550,7 @@ const live: Layer.Layer< async experimental_repairToolCall(failed) { // (a) Tool name case fix only — keep failed.toolCall.input exactly as-is. const lower = failed.toolCall.toolName.toLowerCase() - if (lower !== failed.toolCall.toolName && prepared.tools[lower]) { + if (lower !== failed.toolCall.toolName && runtimeTools[lower]) { l.info("tool call repair: name case fix", { tool: failed.toolCall.toolName, repaired: lower, @@ -465,9 +607,9 @@ const live: Layer.Layer< temperature: prepared.params.temperature, topP: prepared.params.topP, topK: prepared.params.topK, - providerOptions: ProviderTransform.providerOptions(input.model, prepared.params.options), - activeTools: Object.keys(prepared.tools).filter((x) => x !== "invalid"), - tools: prepared.tools, + providerOptions: ProviderTransform.providerOptions(input.model, effectiveOptions ?? {}), + activeTools: Object.keys(runtimeTools).filter((x) => x !== "invalid"), + tools: runtimeTools, toolChoice: effectiveToolChoice, maxOutputTokens: prepared.params.maxOutputTokens, abortSignal: input.abort, diff --git a/packages/deepagent-code/src/session/llm/ai-sdk.ts b/packages/deepagent-code/src/session/llm/ai-sdk.ts index c321d858..7afa39c8 100644 --- a/packages/deepagent-code/src/session/llm/ai-sdk.ts +++ b/packages/deepagent-code/src/session/llm/ai-sdk.ts @@ -2,6 +2,7 @@ import { FinishReason, LLMEvent, ProviderMetadata, ToolResultValue } from "@deep import { Effect, Schema } from "effect" import { type streamText } from "ai" import { errorMessage } from "@/util/error" +import { FreeformTools } from "./freeform-tools" type Result = Awaited> type AISDKEvent = Result["fullStream"] extends AsyncIterable ? T : never @@ -27,6 +28,14 @@ function providerMetadata(value: unknown): ProviderMetadata | undefined { return Schema.is(ProviderMetadata)(value) ? value : undefined } +function toolProviderMetadata(value: unknown, type: "custom" | "function"): ProviderMetadata { + const existing = providerMetadata(value) + return { + ...(existing ?? {}), + deepagent: { ...(existing?.deepagent ?? {}), toolType: type }, + } +} + // Temporary AI SDK bridge: Copilot billing survives only in raw provider chunks here. // Move this extraction into @deepagent-code/llm when Copilot is handled by the native runtime. function copilotTotalNanoAiu(value: unknown) { @@ -220,13 +229,17 @@ export function toLLMEvents( case "tool-call": return Effect.sync(() => { state.toolNames[event.toolCallId] = event.toolName + const custom = event.toolName === "apply_patch" && typeof event.input === "string" return [ LLMEvent.toolCall({ id: event.toolCallId, name: event.toolName, - input: event.input, + input: custom ? FreeformTools.input(event.input as string) : event.input, providerExecuted: "providerExecuted" in event ? event.providerExecuted : undefined, - providerMetadata: providerMetadata(event.providerMetadata), + providerMetadata: + custom || event.toolName === "apply_patch" + ? toolProviderMetadata(event.providerMetadata, custom ? "custom" : "function") + : providerMetadata(event.providerMetadata), }), ] }) diff --git a/packages/deepagent-code/src/session/llm/freeform-tools.ts b/packages/deepagent-code/src/session/llm/freeform-tools.ts new file mode 100644 index 00000000..9bb7c8db --- /dev/null +++ b/packages/deepagent-code/src/session/llm/freeform-tools.ts @@ -0,0 +1,41 @@ +import { openai } from "@ai-sdk/openai" +import type { Tool } from "ai" +import { APPLY_PATCH_LARK_GRAMMAR } from "@/tool/apply-patch-grammar" + +type ProtocolModel = { readonly provider: string } + +export const supportsApplyPatch = (model: ProtocolModel) => model.provider.endsWith(".responses") + +export const format = { + type: "grammar" as const, + syntax: "lark" as const, + definition: APPLY_PATCH_LARK_GRAMMAR, +} + +export const input = (value: unknown) => ({ patchText: typeof value === "string" ? value : "" }) + +export const tools = (model: ProtocolModel, source: Record): Record => { + if (!supportsApplyPatch(model)) { + const { apply_patch: _applyPatch, ...result } = source + return result + } + const patch = source.apply_patch + if (!patch?.execute) { + const { apply_patch: _applyPatch, ...result } = source + return result + } + const execute = patch.execute + const result = { ...source } + result.apply_patch = { + ...openai.tools.customTool({ + name: "apply_patch", + description: `${patch.description ?? "Apply a patch to workspace files."}\nThis is a FREEFORM tool. Send the patch directly without a JSON wrapper.`, + format, + }), + execute: (value: string, options: Parameters[1]) => execute(input(value), options), + toModelOutput: patch.toModelOutput, + } + return result +} + +export * as FreeformTools from "./freeform-tools" diff --git a/packages/deepagent-code/src/session/llm/native-runtime.ts b/packages/deepagent-code/src/session/llm/native-runtime.ts index dedddb64..4f8d04ac 100644 --- a/packages/deepagent-code/src/session/llm/native-runtime.ts +++ b/packages/deepagent-code/src/session/llm/native-runtime.ts @@ -18,6 +18,7 @@ import { } from "@deepagent-code/llm" import type { LLMClientShape } from "@deepagent-code/llm/route" import { LLMNative } from "./native-request" +import { FreeformTools } from "./freeform-tools" export type RuntimeStatus = | { readonly type: "supported"; readonly apiKey: string; readonly baseURL?: string } @@ -127,7 +128,11 @@ export function stream(input: StreamInput): StreamResult { Stream.flatMap((event) => event.type !== "tool-call" || event.providerExecuted ? Stream.make(event) - : Stream.make(event).pipe( + : Stream.make( + event.name === "apply_patch" && typeof event.input === "string" + ? { ...event, input: FreeformTools.input(event.input) } + : event, + ).pipe( Stream.concat( Stream.fromEffectDrain( ToolRuntime.dispatch(tools, event).pipe( @@ -186,7 +191,12 @@ export function nativeTools(tools: Record, input: Pick Effect.tryPromise({ try: () => { diff --git a/packages/deepagent-code/src/tool/apply-patch-grammar.ts b/packages/deepagent-code/src/tool/apply-patch-grammar.ts new file mode 100644 index 00000000..6f6743f6 --- /dev/null +++ b/packages/deepagent-code/src/tool/apply-patch-grammar.ts @@ -0,0 +1,20 @@ +export const APPLY_PATCH_LARK_GRAMMAR = `start: begin_patch hunk+ end_patch +begin_patch: "*** Begin Patch" LF +end_patch: "*** End Patch" LF? + +hunk: add_hunk | delete_hunk | update_hunk +add_hunk: "*** Add File: " filename LF add_line+ +delete_hunk: "*** Delete File: " filename LF +update_hunk: "*** Update File: " filename LF change_move? change? + +filename: /(.+)/ +add_line: "+" /(.*)/ LF -> line + +change_move: "*** Move to: " filename LF +change: (change_context | change_line)+ eof_line? +change_context: ("@@" | "@@ " /(.+)/) LF +change_line: ("+" | "-" | " ") /(.*)/ LF +eof_line: "*** End of File" LF + +%import common.LF +` diff --git a/packages/deepagent-code/src/tool/apply_patch.txt b/packages/deepagent-code/src/tool/apply_patch.txt index 5b2d9560..8d475346 100644 --- a/packages/deepagent-code/src/tool/apply_patch.txt +++ b/packages/deepagent-code/src/tool/apply_patch.txt @@ -31,3 +31,4 @@ It is important to remember: - You must include a header with your intended action (Add/Delete/Update) - You must prefix new lines with `+` even when creating a new file +- If the patch would exceed 12000 UTF-8 bytes (roughly 4000 Chinese characters), use `apply_patch_chunk` instead. Begin at offset 0, then use each result's nextOffset for every append and the final commit. Do not embed a large patch in one JSON tool call. diff --git a/packages/deepagent-code/src/tool/apply_patch_chunk.ts b/packages/deepagent-code/src/tool/apply_patch_chunk.ts new file mode 100644 index 00000000..f54a815d --- /dev/null +++ b/packages/deepagent-code/src/tool/apply_patch_chunk.ts @@ -0,0 +1,146 @@ +import { Effect, Schema } from "effect" +import * as Tool from "./tool" +import { ApplyPatchTool } from "./apply_patch" + +const MAX_CHUNK_BYTES = 12_000 +const MAX_PATCH_BYTES = 2_000_000 +const MAX_TRANSACTIONS_PER_SESSION = 8 +const TRANSACTION_TTL_MS = 30 * 60 * 1000 +const encoder = new TextEncoder() + +interface Metadata { + [key: string]: unknown + transactionID?: string + size?: number + chunks?: number + nextOffset?: number +} + +export const Parameters = Schema.Struct({ + action: Schema.Literals(["begin", "append", "commit", "abort"]).annotate({ + description: "begin starts a transaction; append adds text; commit validates and applies; abort discards it", + }), + transactionID: Schema.optional(Schema.String).annotate({ + description: "The transaction ID returned by begin; required for append, commit, and abort", + }), + offset: Schema.optional(Schema.Int.check(Schema.isGreaterThanOrEqualTo(0))).annotate({ + description: "UTF-8 byte offset for append or commit; must equal nextOffset from the previous result", + }), + patchText: Schema.optional(Schema.String).annotate({ + description: `A verbatim patch chunk of at most ${MAX_CHUNK_BYTES} UTF-8 bytes`, + }), +}) + +export const ApplyPatchChunkTool = Tool.define( + "apply_patch_chunk", + Effect.gen(function* () { + const patch = yield* ApplyPatchTool + const executePatch = yield* Tool.init(patch) + const transactions = new Map< + string, + { + readonly sessionID: string + readonly createdAt: number + readonly chunks: ReadonlyArray + readonly size: number + } + >() + + const fail = (message: string) => Effect.fail(new Error(`apply_patch_chunk failed: ${message}`)) + + const run: ( + params: Schema.Schema.Type, + ctx: Tool.Context, + ) => Effect.Effect, Error> = Effect.fn("ApplyPatchChunkTool.execute")( + function* (params, ctx) { + Array.from(transactions.entries()) + .filter((entry) => Date.now() - entry[1].createdAt > TRANSACTION_TTL_MS) + .forEach((entry) => transactions.delete(entry[0])) + + const patchBytes = params.patchText ? encoder.encode(params.patchText).byteLength : 0 + if (patchBytes > MAX_CHUNK_BYTES) { + return yield* fail( + `patchText has ${patchBytes} UTF-8 bytes; split it into chunks of at most ${MAX_CHUNK_BYTES} bytes`, + ) + } + + if (params.action === "begin") { + if (!params.patchText) return yield* fail("begin requires a non-empty patchText chunk") + if (params.offset !== undefined && params.offset !== 0) return yield* fail("begin offset must be 0") + const active = Array.from(transactions.values()).filter((item) => item.sessionID === ctx.sessionID).length + if (active >= MAX_TRANSACTIONS_PER_SESSION) { + return yield* fail(`session already has ${MAX_TRANSACTIONS_PER_SESSION} active transactions`) + } + const transactionID = `patch_${crypto.randomUUID()}` + transactions.set(transactionID, { + sessionID: ctx.sessionID, + createdAt: Date.now(), + chunks: [params.patchText], + size: patchBytes, + }) + return { + title: "Patch transaction started", + metadata: { transactionID, size: patchBytes, chunks: 1, nextOffset: patchBytes }, + output: `Patch transaction ${transactionID} started with ${patchBytes} UTF-8 bytes. Use offset ${patchBytes} for the next append or commit. No workspace files have changed.`, + } + } + + if (!params.transactionID) return yield* fail(`${params.action} requires transactionID`) + const transaction = transactions.get(params.transactionID) + if (!transaction || transaction.sessionID !== ctx.sessionID) { + return yield* fail(`transaction not found: ${params.transactionID}`) + } + + if (params.action === "abort") { + transactions.delete(params.transactionID) + return { + title: "Patch transaction aborted", + metadata: { transactionID: params.transactionID }, + output: `Patch transaction ${params.transactionID} was discarded. No workspace files changed.`, + } + } + + if (params.offset === undefined) return yield* fail(`${params.action} requires offset`) + if (params.offset !== transaction.size) { + return yield* fail( + `${params.action} offset ${params.offset} does not match next expected UTF-8 byte offset ${transaction.size}`, + ) + } + + if (params.action === "commit" && params.patchText !== undefined) { + return yield* fail("commit does not accept patchText; append the final chunk first") + } + + const size = transaction.size + patchBytes + if (size > MAX_PATCH_BYTES) { + transactions.delete(params.transactionID) + return yield* fail(`assembled patch exceeds ${MAX_PATCH_BYTES} UTF-8 bytes and was discarded`) + } + const chunks = params.patchText ? [...transaction.chunks, params.patchText] : transaction.chunks + + if (params.action === "append") { + if (!params.patchText) return yield* fail("append requires a non-empty patchText chunk") + transactions.set(params.transactionID, { ...transaction, chunks, size }) + return { + title: "Patch chunk staged", + metadata: { transactionID: params.transactionID, size, chunks: chunks.length, nextOffset: size }, + output: `Staged chunk ${chunks.length} for ${params.transactionID}; ${size} UTF-8 bytes total. Use offset ${size} for the next append or commit. No workspace files have changed.`, + } + } + + transactions.delete(params.transactionID) + const result = yield* executePatch.execute({ patchText: chunks.join("") }, ctx) + return { ...result, metadata: result.metadata as Metadata } + }, + ) + + return { + description: `Stage a large apply_patch payload in bounded JSON chunks, then validate and apply it as one transaction. + +Use this only when a normal apply_patch call may be too large for one tool call. Call begin with offset 0 and the first patchText chunk. For every append and the final commit, send the exact nextOffset returned by the previous result. Chunks are concatenated verbatim. Each patchText must be at most ${MAX_CHUNK_BYTES} UTF-8 bytes; keep Chinese-language chunks below roughly 4000 characters. Commit never accepts patchText. No workspace file is changed before commit. Abort discards the staged patch.`, + parameters: Parameters, + execute: (params: Schema.Schema.Type, ctx: Tool.Context) => + run(params, ctx).pipe(Effect.orDie), + } satisfies Tool.DefWithoutID + }), +) diff --git a/packages/deepagent-code/src/tool/edit.txt b/packages/deepagent-code/src/tool/edit.txt index 618fd5ad..dad560c9 100644 --- a/packages/deepagent-code/src/tool/edit.txt +++ b/packages/deepagent-code/src/tool/edit.txt @@ -4,6 +4,7 @@ Usage: - You must use your `Read` tool at least once in the conversation before editing. This tool will error if you attempt an edit without reading the file. - When editing text from Read tool output, ensure you preserve the exact indentation (tabs/spaces) as it appears AFTER the line number prefix. The line number prefix format is: line number + colon + space (e.g., `1: `). Everything after that space is the actual file content to match. Never include any part of the line number prefix in the oldString or newString. - ALWAYS prefer editing existing files in the codebase. NEVER write new files unless explicitly required. +- Keep oldString and newString as small as possible. If the replacement would make this tool call large, use `apply_patch_chunk`, begin at offset 0, and use each result's nextOffset for every append and the final commit. - Only use emojis if the user explicitly requests it. Avoid adding emojis to files unless asked. - The edit will FAIL if `oldString` is not found in the file with an error "oldString not found in content". - The edit will FAIL if `oldString` is found multiple times in the file with an error "Found multiple matches for oldString. Provide more surrounding lines in oldString to identify the correct match." Either provide a larger string with more surrounding context to make it unique or use `replaceAll` to change every instance of `oldString`. diff --git a/packages/deepagent-code/src/tool/registry.ts b/packages/deepagent-code/src/tool/registry.ts index 7e22104b..a3053c98 100644 --- a/packages/deepagent-code/src/tool/registry.ts +++ b/packages/deepagent-code/src/tool/registry.ts @@ -39,6 +39,7 @@ import { InstanceStore } from "@/project/instance-store" import { InstanceBootstrap } from "@/project/bootstrap-service" import * as Truncate from "./truncate" import { ApplyPatchTool } from "./apply_patch" +import { ApplyPatchChunkTool } from "./apply_patch_chunk" import { Glob } from "@deepagent-code/core/util/glob" import path from "path" import { pathToFileURL } from "url" @@ -145,6 +146,7 @@ export const layer: Layer.Layer< const edit = yield* EditTool const greptool = yield* GrepTool const patchtool = yield* ApplyPatchTool + const patchchunk = yield* ApplyPatchChunkTool const skilltool = yield* SkillTool const codeintel = yield* CodeIntelTool const profiletool = yield* ProfileTool @@ -267,6 +269,7 @@ export const layer: Layer.Layer< search: Tool.init(websearch), skill: Tool.init(skilltool), patch: Tool.init(patchtool), + patch_chunk: Tool.init(patchchunk), question: Tool.init(question), lsp: Tool.init(lsptool), code_intel: Tool.init(codeintel), @@ -296,6 +299,7 @@ export const layer: Layer.Layer< tool.search, tool.skill, tool.patch, + tool.patch_chunk, tool.planwrite, ...(flags.experimentalLspTool ? [tool.lsp] : []), ...(flags.codeIntelTool ? [tool.code_intel] : []), @@ -340,11 +344,6 @@ export const layer: Layer.Layer< return webSearchEnabled(input.providerID, { exa: flags.enableExa, parallel: flags.enableParallel }) } - const usePatch = - input.modelID.includes("gpt-") && !input.modelID.includes("oss") && !input.modelID.includes("gpt-4") - if (tool.id === ApplyPatchTool.id) return usePatch - if (tool.id === EditTool.id || tool.id === WriteTool.id) return !usePatch - return true }) diff --git a/packages/deepagent-code/src/tool/write.txt b/packages/deepagent-code/src/tool/write.txt index 063cbb1f..a21cfe3f 100644 --- a/packages/deepagent-code/src/tool/write.txt +++ b/packages/deepagent-code/src/tool/write.txt @@ -4,5 +4,6 @@ Usage: - This tool will overwrite the existing file if there is one at the provided path. - If this is an existing file, you MUST use the Read tool first to read the file's contents. This tool will fail if you did not read the file first. - ALWAYS prefer editing existing files in the codebase. NEVER write new files unless explicitly required. +- If the file content would make this tool call large, use `apply_patch_chunk` instead. Begin at offset 0, use each result's nextOffset for every append and the final commit, and keep each chunk at most 12000 UTF-8 bytes. Do not embed a large file or patch in one JSON tool call. - NEVER proactively create documentation files (*.md) or README files. Only create documentation files if explicitly requested by the User. - Only use emojis if the user explicitly requests it. Avoid writing emojis to files unless asked. diff --git a/packages/deepagent-code/test/session/llm-native.test.ts b/packages/deepagent-code/test/session/llm-native.test.ts index 1b59af21..2310323a 100644 --- a/packages/deepagent-code/test/session/llm-native.test.ts +++ b/packages/deepagent-code/test/session/llm-native.test.ts @@ -1,10 +1,11 @@ import { describe, expect, test } from "bun:test" -import { LLMEvent, ToolFailure } from "@deepagent-code/llm" +import { LLMEvent, ToolFailure, toDefinitions } from "@deepagent-code/llm" import { LLMClient, RequestExecutor, WebSocketExecutor, type LLMClientShape } from "@deepagent-code/llm/route" import { jsonSchema, tool, type ModelMessage, type Tool } from "ai" import { Effect, Fiber, Layer, Stream } from "effect" import { LLMNative } from "@/session/llm/native-request" import { LLMNativeRuntime } from "@/session/llm/native-runtime" +import { FreeformTools } from "@/session/llm/freeform-tools" import type { Provider } from "@/provider/provider" import { OAUTH_DUMMY_KEY } from "@/auth" @@ -520,6 +521,81 @@ describe("session.llm-native.request", () => { }), ) + it.effect("adapts apply_patch to a raw custom tool while preserving its executor", () => + Effect.gen(function* () { + const calls: unknown[] = [] + const source = FreeformTools.tools( + { provider: "openai.responses" }, + { + apply_patch: tool({ + description: "Apply a patch", + inputSchema: jsonSchema({ + type: "object", + properties: { patchText: { type: "string" } }, + required: ["patchText"], + }), + execute: async (value) => { + calls.push(value) + return { output: "done" } + }, + }), + apply_patch_chunk: tool({ + description: "Chunk fallback", + inputSchema: jsonSchema({ type: "object" }), + }), + }, + ) + + expect(source.apply_patch.type).toBe("provider") + expect(source.apply_patch_chunk).toBeDefined() + const wrapped = LLMNativeRuntime.nativeTools(source, { + messages: [], + abort: new AbortController().signal, + }) + expect(toDefinitions(wrapped)[0]).toMatchObject({ + name: "apply_patch", + type: "custom", + format: { type: "grammar", syntax: "lark" }, + }) + + yield* wrapped.apply_patch.execute("*** Begin Patch\n*** End Patch", { + id: "call_patch", + name: "apply_patch", + }) + expect(calls).toEqual([{ patchText: "*** Begin Patch\n*** End Patch" }]) + }), + ) + + it.effect("keeps JSON file tools for Chat and Anthropic while removing raw apply_patch", () => + Effect.gen(function* () { + const source = { + read: tool({ description: "Read", inputSchema: jsonSchema({ type: "object" }) }), + edit: tool({ description: "Edit", inputSchema: jsonSchema({ type: "object" }) }), + write: tool({ description: "Write", inputSchema: jsonSchema({ type: "object" }) }), + apply_patch: tool({ + description: "Patch", + inputSchema: jsonSchema({ type: "object" }), + execute: async () => ({ output: "patched" }), + }), + apply_patch_chunk: tool({ description: "Chunk", inputSchema: jsonSchema({ type: "object" }) }), + } + + expect(Object.keys(FreeformTools.tools({ provider: "openai.chat" }, source))).toEqual([ + "read", + "edit", + "write", + "apply_patch_chunk", + ]) + expect(Object.keys(FreeformTools.tools({ provider: "anthropic.messages" }, source))).toEqual([ + "read", + "edit", + "write", + "apply_patch_chunk", + ]) + expect(FreeformTools.tools({ provider: "openai.responses" }, source).apply_patch.type).toBe("provider") + }), + ) + it.effect("native tool wrapper raises ToolFailure when the source tool has no execute handler", () => Effect.gen(function* () { // The AI SDK Tool shape allows execute to be omitted (e.g., client-side / MCP tools). diff --git a/packages/deepagent-code/test/session/llm.test.ts b/packages/deepagent-code/test/session/llm.test.ts index d795cb61..dba6814f 100644 --- a/packages/deepagent-code/test/session/llm.test.ts +++ b/packages/deepagent-code/test/session/llm.test.ts @@ -240,6 +240,110 @@ describe("session.llm.hasToolCalls", () => { }) }) +describe("session.llm.decideToolChoice", () => { + const model = (npm: string, id = "test-model", options: { toolcall?: boolean; reasoning?: boolean } = {}) => + ({ + id: ModelV2.ID.make(id), + providerID: ProviderV2.ID.make("test"), + api: { id, url: "https://example.test", npm }, + name: id, + capabilities: { + toolcall: options.toolcall ?? true, + reasoning: options.reasoning ?? false, + attachment: false, + temperature: true, + input: { text: true, audio: false, image: false, video: false, pdf: false }, + output: { text: true, audio: false, image: false, video: false, pdf: false }, + interleaved: false, + }, + cost: { input: 0, output: 0, cache: { read: 0, write: 0 } }, + limit: { context: 100_000, input: 100_000, output: 10_000 }, + status: "active", + options: {}, + headers: {}, + release_date: "2026-01-01", + }) satisfies Provider.Model + + test("fails closed instead of silently relaxing required under thinking", () => { + expect(LLM.decideToolChoice("required", { reasoningEffort: "high" })).toEqual({ + capability: "unsupported_thinking_with_forced_tool", + }) + }) + + test("allows required after the caller disables thinking", () => { + expect(LLM.decideToolChoice("required", {})).toEqual({ + capability: "forced_tool", + toolChoice: "required", + }) + }) + + test("declares the bounded auto-only fallback", () => { + expect(LLM.decideToolChoice("auto", { reasoningEffort: "high" })).toEqual({ + capability: "auto_only", + toolChoice: "auto", + }) + }) + + test.each([ + ["@ai-sdk/openai", "openai_responses"], + ["@ai-sdk/openai-compatible", "openai_chat"], + ["@ai-sdk/anthropic", "anthropic_messages"], + ["@ai-sdk/google", "gemini"], + ["@ai-sdk/amazon-bedrock", "bedrock_converse"], + ] as const)("uses forced finalization for verified %s protocol", (npm, protocol) => { + expect(LLM.finalizerCapability(model(npm))).toEqual({ + capability: "forced_tool", + protocol, + reasoning: "disabled", + toolChoice: "required", + }) + }) + + test("reasoning-only models use bounded auto without an unsupported forced combination", () => { + expect(LLM.finalizerCapability(model("@ai-sdk/openai", "o3", { reasoning: true }))).toEqual({ + capability: "auto_only", + protocol: "openai_responses", + reasoning: "inherit", + toolChoice: "auto", + reason: "reasoning_cannot_be_disabled", + }) + }) + + test("unknown protocols use bounded auto and ordinary required callers fail closed", () => { + const custom = model("file:///custom-provider.ts") + expect(LLM.finalizerCapability(custom)).toMatchObject({ + capability: "auto_only", + protocol: "unknown", + reason: "protocol_forced_tool_unverified", + }) + expect(LLM.decideToolChoice("required", {}, custom)).toEqual({ capability: "unsupported_forced_tool" }) + }) + + test("models without tool calls reject structured finalization before provider execution", () => { + expect(LLM.finalizerCapability(model("@ai-sdk/openai", "text-only", { toolcall: false }))).toMatchObject({ + capability: "unsupported", + reason: "model_has_no_tool_call_capability", + }) + }) + + test("per-turn disable removes merged reasoning options without mutating defaults", () => { + const options = { + reasoningEffort: "high", + thinking: { type: "enabled" }, + thinkingConfig: { thinkingBudget: 16_000 }, + enable_thinking: true, + chat_template_args: { enable_thinking: true, other: "kept" }, + modelParams: { reasoning: { effort: "high" }, temperature: 0 }, + } + expect(LLM.disableThinking(options)).toEqual({ + enable_thinking: false, + chat_template_args: { enable_thinking: false, other: "kept" }, + modelParams: { temperature: 0 }, + }) + expect(options.reasoningEffort).toBe("high") + }) +}) + describe("session.llm.ai-sdk adapter", () => { type AISDKAdapterEvent = Parameters[1] @@ -388,6 +492,28 @@ describe("session.llm.ai-sdk adapter", () => { ]) }) + test("maps raw apply_patch input to the canonical executor shape", async () => { + const patch = "*** Begin Patch\n*** End Patch" + const events = await adapt([ + uncheckedAdapterEvent({ + type: "tool-call", + toolCallId: "call-patch", + toolName: "apply_patch", + input: patch, + }), + ]) + + expect(events).toEqual([ + { + type: "tool-call", + id: "call-patch", + name: "apply_patch", + input: { patchText: patch }, + providerMetadata: { deepagent: { toolType: "custom" } }, + }, + ]) + }) + test("explicitly ignores non-session-visible AI SDK chunks", async () => { expect( await adapt([ @@ -1333,9 +1459,7 @@ describe("session.llm.stream", () => { }), }), ) - const injectedClient = LLMClient.layer.pipe( - Layer.provide(Layer.mergeAll(executor, WebSocketExecutor.layer)), - ) + const injectedClient = LLMClient.layer.pipe(Layer.provide(Layer.mergeAll(executor, WebSocketExecutor.layer))) const sessionID = SessionID.make("session-test-native-injected-tool") const agent = { name: "test", diff --git a/packages/deepagent-code/test/tool/apply_patch_chunk.test.ts b/packages/deepagent-code/test/tool/apply_patch_chunk.test.ts new file mode 100644 index 00000000..98af07a9 --- /dev/null +++ b/packages/deepagent-code/test/tool/apply_patch_chunk.test.ts @@ -0,0 +1,131 @@ +import { describe, expect } from "bun:test" +import path from "path" +import { Cause, Effect, Exit, Layer } from "effect" +import { Agent } from "@/agent/agent" +import { EventV2Bridge } from "@/event-v2-bridge" +import { Format } from "@/format" +import { LSP } from "@/lsp/lsp" +import { MessageID, SessionID } from "@/session/schema" +import { ApplyPatchChunkTool } from "@/tool/apply_patch_chunk" +import * as Tool from "@/tool/tool" +import { Truncate } from "@/tool/truncate" +import { FSUtil } from "@deepagent-code/core/fs-util" +import { TestInstance } from "../fixture/fixture" +import { testEffect } from "../lib/effect" + +const it = testEffect( + Layer.mergeAll( + LSP.defaultLayer, + FSUtil.defaultLayer, + Format.defaultLayer, + EventV2Bridge.defaultLayer, + Truncate.defaultLayer, + Agent.defaultLayer, + ), +) + +const context = (sessionID = SessionID.make("ses_patch_chunk")): Tool.Context => ({ + sessionID, + messageID: MessageID.make("msg_patch_chunk"), + callID: "call_patch_chunk", + agent: "build", + abort: AbortSignal.any([]), + messages: [], + metadata: () => Effect.void, + ask: () => Effect.void, +}) + +const failure = (effect: Effect.Effect, message: string) => + Effect.gen(function* () { + const exit = yield* Effect.exit(effect) + expect(Exit.isFailure(exit)).toBe(true) + if (Exit.isFailure(exit)) expect(Cause.pretty(exit.cause)).toContain(message) + }) + +describe("tool.apply_patch_chunk", () => { + it.instance("stages chunks without touching the workspace and applies only on commit", () => + Effect.gen(function* () { + const test = yield* TestInstance + const info = yield* ApplyPatchChunkTool + const tool = yield* info.init() + const target = path.join(test.directory, "chunked.txt") + const patch = "*** Begin Patch\n*** Add File: chunked.txt\n+chunked content\n*** End Patch" + const first = yield* tool.execute({ action: "begin", patchText: patch.slice(0, 32) }, context()) + const transactionID = first.metadata.transactionID + + expect(transactionID).toBeString() + expect(first.metadata.nextOffset).toBe(32) + expect(yield* Effect.promise(() => Bun.file(target).exists())).toBe(false) + + const appended = yield* tool.execute( + { action: "append", transactionID: String(transactionID), offset: 32, patchText: patch.slice(32) }, + context(), + ) + expect(appended.metadata.nextOffset).toBe(new TextEncoder().encode(patch).byteLength) + expect(yield* Effect.promise(() => Bun.file(target).exists())).toBe(false) + + yield* tool.execute( + { action: "commit", transactionID: String(transactionID), offset: Number(appended.metadata.nextOffset) }, + context(), + ) + expect(yield* Effect.promise(() => Bun.file(target).text())).toBe("chunked content\n") + }), + ) + + it.instance("rejects oversized chunks without creating a transaction", () => + Effect.gen(function* () { + const info = yield* ApplyPatchChunkTool + const tool = yield* info.init() + + yield* failure( + tool.execute({ action: "begin", patchText: "x".repeat(12_001) }, context()), + "split it into chunks", + ) + }), + ) + + it.instance("does not allow another session to append or commit a transaction", () => + Effect.gen(function* () { + const info = yield* ApplyPatchChunkTool + const tool = yield* info.init() + const started = yield* tool.execute( + { action: "begin", patchText: "*** Begin Patch\n" }, + context(SessionID.make("ses_owner")), + ) + + yield* failure( + tool.execute( + { action: "append", transactionID: String(started.metadata.transactionID), patchText: "*** End Patch" }, + context(SessionID.make("ses_other")), + ), + "transaction not found", + ) + }), + ) + + it.instance("rejects missing, duplicate, and out-of-order offsets without changing files", () => + Effect.gen(function* () { + const test = yield* TestInstance + const info = yield* ApplyPatchChunkTool + const tool = yield* info.init() + const target = path.join(test.directory, "ordered.txt") + const started = yield* tool.execute({ action: "begin", offset: 0, patchText: "*** Begin Patch\n" }, context()) + const transactionID = String(started.metadata.transactionID) + const nextOffset = Number(started.metadata.nextOffset) + + yield* failure( + tool.execute({ action: "append", transactionID, patchText: "*** End Patch" }, context()), + "append requires offset", + ) + yield* failure( + tool.execute({ action: "append", transactionID, offset: 0, patchText: "*** End Patch" }, context()), + "does not match next expected", + ) + yield* failure( + tool.execute({ action: "commit", transactionID, offset: nextOffset + 1 }, context()), + "does not match next expected", + ) + expect(yield* Effect.promise(() => Bun.file(target).exists())).toBe(false) + }), + ) +}) diff --git a/packages/llm/src/cache-policy.ts b/packages/llm/src/cache-policy.ts index 60f96dc6..2bde1982 100644 --- a/packages/llm/src/cache-policy.ts +++ b/packages/llm/src/cache-policy.ts @@ -48,7 +48,7 @@ const markLastTool = (tools: ReadonlyArray, hint: CacheHint): Re if (tools.length === 0) return tools const last = tools.length - 1 if (tools[last]!.cache) return tools - return tools.map((tool, i) => (i === last ? new ToolDefinition({ ...tool, cache: hint }) : tool)) + return tools.map((tool, i) => (i === last ? ToolDefinition.make({ ...tool, cache: hint }) : tool)) } const markLastSystem = (system: LLMRequest["system"], hint: CacheHint): LLMRequest["system"] => { diff --git a/packages/llm/src/protocols/anthropic-messages.ts b/packages/llm/src/protocols/anthropic-messages.ts index 5331d68b..907d42a9 100644 --- a/packages/llm/src/protocols/anthropic-messages.ts +++ b/packages/llm/src/protocols/anthropic-messages.ts @@ -6,14 +6,15 @@ import { Framing } from "../route/framing" import { Protocol } from "../route/protocol" import { LLMEvent, + ToolDefinition, Usage, + type FunctionToolDefinition, type CacheHint, type FinishReason, type LLMRequest, type MediaPart, type ProviderMetadata, type ToolCallPart, - type ToolDefinition, type ToolResultContentPart, type ToolResultPart, } from "../schema" @@ -220,6 +221,7 @@ type AnthropicEvent = Schema.Schema.Type interface ParserState { readonly tools: ToolStream.State + readonly pendingToolEvents: ReadonlyArray readonly usage?: Usage readonly lifecycle: Lifecycle.State } @@ -256,7 +258,7 @@ const signatureFromMetadata = (metadata: ProviderMetadata | undefined): string | return typeof anthropic.signature === "string" ? anthropic.signature : undefined } -const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition): AnthropicTool => ({ +const lowerTool = (breakpoints: Cache.Breakpoints, tool: FunctionToolDefinition): AnthropicTool => ({ name: tool.name, description: tool.description, input_schema: tool.inputSchema, @@ -502,6 +504,8 @@ const lowerThinking = Effect.fn("AnthropicMessages.lowerThinking")(function* (re }) const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) { + if (!request.tools.every(ToolDefinition.isFunction)) + return yield* invalid("Anthropic Messages does not support custom text tools") const toolChoice = request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined const generation = request.generation // Allocate the 4-breakpoint budget in invalidation order: tools → system → @@ -756,30 +760,43 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun ) { if (event.index === undefined) return [state, NO_EVENTS] satisfies StepResult const result = yield* ToolStream.finish(ADAPTER, state.tools, event.index) - const events: LLMEvent[] = [] const resultEvents = result.events ?? [] - const lifecycle = resultEvents.length - ? Lifecycle.stepStart(state.lifecycle, events) - : Lifecycle.reasoningEnd( - Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`), - events, - `reasoning-${event.index}`, - ) - events.push(...resultEvents) + if (resultEvents.some((item) => item.type === "tool-call" && item.providerExecuted === true)) { + const events: LLMEvent[] = [] + const lifecycle = Lifecycle.stepStart(state.lifecycle, events) + events.push(...resultEvents) + return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult + } + if (resultEvents.length) + return [ + { ...state, tools: result.tools, pendingToolEvents: [...state.pendingToolEvents, ...resultEvents] }, + NO_EVENTS, + ] satisfies StepResult + const events: LLMEvent[] = [] + const lifecycle = Lifecycle.reasoningEnd( + Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`), + events, + `reasoning-${event.index}`, + ) return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult }) const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult => { const usage = mergeUsage(state.usage, mapUsage(event.usage)) - const events: LLMEvent[] = [] + const reason = mapFinishReason(event.delta?.stop_reason) + const releaseTools = reason === "stop" || reason === "tool-calls" + const events: LLMEvent[] = releaseTools ? [...state.pendingToolEvents] : [] + const hasLocalToolCall = state.pendingToolEvents.some( + (item) => item.type === "tool-call" && item.providerExecuted !== true, + ) const lifecycle = Lifecycle.finish(state.lifecycle, events, { - reason: mapFinishReason(event.delta?.stop_reason), + reason: releaseTools && hasLocalToolCall ? "tool-calls" : reason, usage, providerMetadata: event.delta?.stop_sequence ? anthropicMetadata({ stopSequence: event.delta.stop_sequence }) : undefined, }) - return [{ ...state, lifecycle, usage }, events] + return [{ ...state, lifecycle, pendingToolEvents: [], usage }, events] } // Prefix `error.type` so overloads, rate limits, and quota errors are visible @@ -827,7 +844,7 @@ export const protocol = Protocol.make({ }, stream: { event: Protocol.jsonEvent(AnthropicEvent), - initial: () => ({ tools: ToolStream.empty(), lifecycle: Lifecycle.initial() }), + initial: () => ({ tools: ToolStream.empty(), pendingToolEvents: [], lifecycle: Lifecycle.initial() }), step, }, }) diff --git a/packages/llm/src/protocols/bedrock-converse.ts b/packages/llm/src/protocols/bedrock-converse.ts index 72143a48..e17e039e 100644 --- a/packages/llm/src/protocols/bedrock-converse.ts +++ b/packages/llm/src/protocols/bedrock-converse.ts @@ -4,14 +4,15 @@ import { Endpoint } from "../route/endpoint" import { Protocol } from "../route/protocol" import { LLMEvent, + ToolDefinition, Usage, type CacheHint, type FinishReason, + type FunctionToolDefinition, type LLMRequest, type ProviderMetadata, type ReasoningPart, type ToolCallPart, - type ToolDefinition, type ToolResultPart, } from "../schema" import { BedrockEventStream } from "./bedrock-event-stream" @@ -205,7 +206,7 @@ type BedrockEvent = Schema.Schema.Type // ============================================================================= // Request Lowering // ============================================================================= -const lowerToolSpec = (tool: ToolDefinition): BedrockToolSpec => ({ +const lowerToolSpec = (tool: FunctionToolDefinition): BedrockToolSpec => ({ toolSpec: { name: tool.name, description: tool.description, @@ -213,7 +214,10 @@ const lowerToolSpec = (tool: ToolDefinition): BedrockToolSpec => ({ }, }) -const lowerTools = (breakpoints: BedrockCache.Breakpoints, tools: ReadonlyArray): BedrockTool[] => { +const lowerTools = ( + breakpoints: BedrockCache.Breakpoints, + tools: ReadonlyArray, +): BedrockTool[] => { const result: BedrockTool[] = [] for (const tool of tools) { result.push(lowerToolSpec(tool)) @@ -374,6 +378,8 @@ const lowerSystem = ( ): BedrockSystemBlock[] => system.flatMap((part) => textWithCache(breakpoints, part.text, part.cache)) const fromRequest = Effect.fn("BedrockConverse.fromRequest")(function* (request: LLMRequest) { + if (!request.tools.every(ToolDefinition.isFunction)) + return yield* ProviderShared.invalidRequest("Bedrock Converse does not support custom text tools") const toolChoice = request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined const generation = request.generation // Bedrock-Claude shares Anthropic's 4-breakpoint cap. Spend the budget in diff --git a/packages/llm/src/protocols/gemini.ts b/packages/llm/src/protocols/gemini.ts index 1db9d43e..51fbee26 100644 --- a/packages/llm/src/protocols/gemini.ts +++ b/packages/llm/src/protocols/gemini.ts @@ -6,14 +6,15 @@ import { Framing } from "../route/framing" import { Protocol } from "../route/protocol" import { LLMEvent, + ToolDefinition, Usage, + type FunctionToolDefinition, type FinishReason, type LLMRequest, type MediaPart, type ProviderMetadata, type TextPart, type ToolCallPart, - type ToolDefinition, type ToolResultContentPart, } from "../schema" import { JsonObject, optionalArray, ProviderShared } from "./shared" @@ -166,7 +167,7 @@ interface ParserState { // ============================================================================= // Request Lowering // ============================================================================= -const lowerTool = (tool: ToolDefinition) => ({ +const lowerTool = (tool: FunctionToolDefinition) => ({ name: tool.name, description: tool.description, parameters: GeminiToolSchema.convert(tool.inputSchema), @@ -298,6 +299,8 @@ const thinkingConfig = (request: LLMRequest) => { } const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMRequest) { + if (!request.tools.every(ToolDefinition.isFunction)) + return yield* ProviderShared.invalidRequest("Gemini does not support custom text tools") const toolsEnabled = request.tools.length > 0 && request.toolChoice?.type !== "none" const generation = request.generation const generationConfig = { diff --git a/packages/llm/src/protocols/openai-chat.ts b/packages/llm/src/protocols/openai-chat.ts index 65c41031..04a15ae6 100644 --- a/packages/llm/src/protocols/openai-chat.ts +++ b/packages/llm/src/protocols/openai-chat.ts @@ -6,14 +6,15 @@ import { HttpTransport } from "../route/transport" import { Protocol } from "../route/protocol" import { LLMEvent, + ToolDefinition, Usage, + type FunctionToolDefinition, type FinishReason, type LLMRequest, type MediaPart, type ReasoningPart, type TextPart, type ToolCallPart, - type ToolDefinition, type ToolResultContentPart, } from "../schema" import { isRecord, JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared" @@ -174,7 +175,7 @@ const invalid = ProviderShared.invalidRequest // Lowering is the only place that knows how common LLM messages map onto the // OpenAI Chat wire format. Keep provider quirks here instead of leaking native // fields into `LLMRequest`. -const lowerTool = (tool: ToolDefinition): OpenAIChatTool => ({ +const lowerTool = (tool: FunctionToolDefinition): OpenAIChatTool => ({ type: "function", function: { name: tool.name, @@ -340,6 +341,8 @@ const lowerOptions = Effect.fn("OpenAIChat.lowerOptions")(function* (request: LL }) const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (request: LLMRequest) { + if (!request.tools.every(ToolDefinition.isFunction)) + return yield* invalid("OpenAI Chat does not support custom text tools") // `fromRequest` returns the provider body only. Endpoint, auth, framing, // validation, and HTTP execution are composed by `Route.make`. const generation = request.generation @@ -430,7 +433,9 @@ const step = (state: ParserState, event: OpenAIChatEvent) => // Finalize accumulated tool inputs eagerly when finish_reason arrives so // JSON parse failures fail the stream at the boundary rather than at halt. const finished = - finishReason !== undefined && state.finishReason === undefined && Object.keys(tools).length > 0 + (finishReason === "stop" || finishReason === "tool-calls") && + state.finishReason === undefined && + Object.keys(tools).length > 0 ? yield* ToolStream.finishAll(ADAPTER, tools) : undefined diff --git a/packages/llm/src/protocols/openai-responses.ts b/packages/llm/src/protocols/openai-responses.ts index 3616ef6c..42e0bbe2 100644 --- a/packages/llm/src/protocols/openai-responses.ts +++ b/packages/llm/src/protocols/openai-responses.ts @@ -6,6 +6,7 @@ import { HttpTransport, WebSocketTransport } from "../route/transport" import { Protocol } from "../route/protocol" import { LLMEvent, + ToolDefinition, Usage, type FinishReason, type LLMRequest, @@ -13,7 +14,6 @@ import { type ReasoningPart, type TextPart, type ToolCallPart, - type ToolDefinition, type ToolResultContentPart, type ToolResultPart, } from "../schema" @@ -85,11 +85,22 @@ const OpenAIResponsesInputItem = Schema.Union([ name: Schema.String, arguments: Schema.String, }), + Schema.Struct({ + type: Schema.tag("custom_tool_call"), + call_id: Schema.String, + name: Schema.String, + input: Schema.String, + }), Schema.Struct({ type: Schema.tag("function_call_output"), call_id: Schema.String, output: OpenAIResponsesFunctionCallOutput, }), + Schema.Struct({ + type: Schema.tag("custom_tool_call_output"), + call_id: Schema.String, + output: OpenAIResponsesFunctionCallOutput, + }), ]) type OpenAIResponsesInputItem = Schema.Schema.Type @@ -102,18 +113,36 @@ type OpenAIResponsesReasoningInput = { encrypted_content?: string | null } -const OpenAIResponsesTool = Schema.Struct({ - type: Schema.tag("function"), - name: Schema.String, - description: Schema.String, - parameters: JsonObject, - strict: Schema.optional(Schema.Boolean), -}) +const OpenAIResponsesTool = Schema.Union([ + Schema.Struct({ + type: Schema.tag("function"), + name: Schema.String, + description: Schema.String, + parameters: JsonObject, + strict: Schema.optional(Schema.Boolean), + }), + Schema.Struct({ + type: Schema.tag("custom"), + name: Schema.String, + description: Schema.optional(Schema.String), + format: Schema.optional( + Schema.Union([ + Schema.Struct({ + type: Schema.Literal("grammar"), + syntax: Schema.Literals(["regex", "lark"]), + definition: Schema.String, + }), + Schema.Struct({ type: Schema.Literal("text") }), + ]), + ), + }), +]) type OpenAIResponsesTool = Schema.Schema.Type const OpenAIResponsesToolChoice = Schema.Union([ Schema.Literals(["auto", "none", "required"]), Schema.Struct({ type: Schema.tag("function"), name: Schema.String }), + Schema.Struct({ type: Schema.tag("custom"), name: Schema.String }), ]) // Fields shared between the HTTP body and the WebSocket `response.create` @@ -177,6 +206,7 @@ const OpenAIResponsesStreamItem = Schema.Struct({ call_id: Schema.optional(Schema.String), name: Schema.optional(Schema.String), arguments: Schema.optional(Schema.String), + input: Schema.optional(Schema.String), // Hosted (provider-executed) tool fields. Each hosted tool item carries its // own subset of these — we capture them generically so we can surface the // call's typed input portion and round-trip the full result payload without @@ -232,6 +262,7 @@ type OpenAIResponsesEvent = Schema.Schema.Type interface ParserState { readonly tools: ToolStream.State + readonly pendingToolEvents: ReadonlyArray readonly hasFunctionCall: boolean readonly lifecycle: Lifecycle.State readonly reasoningItems: Readonly> @@ -253,27 +284,60 @@ const invalid = ProviderShared.invalidRequest // ============================================================================= // Request Lowering // ============================================================================= -const lowerTool = (tool: ToolDefinition): OpenAIResponsesTool => ({ - type: "function", - name: tool.name, - description: tool.description, - parameters: ProviderShared.openAiToolInputSchema(tool.inputSchema), -}) +const lowerTool = (tool: ToolDefinition): OpenAIResponsesTool => + ToolDefinition.isCustom(tool) + ? { + type: "custom", + name: tool.name, + description: tool.description || undefined, + format: tool.format, + } + : { + type: "function", + name: tool.name, + description: tool.description, + parameters: ProviderShared.openAiToolInputSchema(tool.inputSchema), + } -const lowerToolChoice = (toolChoice: NonNullable) => +const lowerToolChoice = (toolChoice: NonNullable, tools: ReadonlyArray) => ProviderShared.matchToolChoice("OpenAI Responses", toolChoice, { auto: () => "auto" as const, none: () => "none" as const, required: () => "required" as const, - tool: (name) => ({ type: "function" as const, name }), + tool: (name) => ({ + type: tools.some((tool) => tool.name === name && ToolDefinition.isCustom(tool)) + ? ("custom" as const) + : ("function" as const), + name, + }), }) -const lowerToolCall = (part: ToolCallPart): OpenAIResponsesInputItem => ({ - type: "function_call", - call_id: part.id, - name: part.name, - arguments: ProviderShared.encodeJson(part.input), -}) +const rawCustomToolInput = (input: unknown) => { + if (typeof input === "string") return input + if (ProviderShared.isRecord(input) && typeof input.patchText === "string") return input.patchText + if (ProviderShared.isRecord(input) && typeof input.value === "string") return input.value + return ProviderShared.encodeJson(input) +} + +const historicalToolType = (part: ToolCallPart | ToolResultPart): "custom" | "function" | undefined => { + const marker = part.providerMetadata?.deepagent?.toolType + return marker === "custom" || marker === "function" ? marker : undefined +} + +const lowerToolCall = (part: ToolCallPart, custom: boolean): OpenAIResponsesInputItem => + custom + ? { + type: "custom_tool_call", + call_id: part.id, + name: part.name, + input: rawCustomToolInput(part.input), + } + : { + type: "function_call", + call_id: part.id, + name: part.name, + arguments: ProviderShared.encodeJson(part.input), + } const lowerReasoning = (part: ReasoningPart): OpenAIResponsesReasoningInput | undefined => { const openai = part.providerMetadata?.openai @@ -339,6 +403,9 @@ const lowerToolResultOutput = Effect.fn("OpenAIResponses.lowerToolResultOutput") }) const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (request: LLMRequest) { + const customTools = new Set(request.tools.filter(ToolDefinition.isCustom).map((tool) => tool.name)) + const customToolCallIDs = new Set() + const functionToolCallIDs = new Set() const system: OpenAIResponsesInputItem[] = request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }] const input: OpenAIResponsesInputItem[] = [...system] @@ -400,7 +467,11 @@ const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (requ if (part.type === "tool-call") { flushText() if (part.providerExecuted === true) continue - input.push(lowerToolCall(part)) + const historicalType = historicalToolType(part) + const isCustom = historicalType === "custom" || (historicalType === undefined && customTools.has(part.name)) + if (isCustom) customToolCallIDs.add(part.id) + else functionToolCallIDs.add(part.id) + input.push(lowerToolCall(part, isCustom)) continue } if (part.type === "tool-result" && part.providerExecuted === true) { @@ -425,11 +496,22 @@ const lowerMessages = Effect.fn("OpenAIResponses.lowerMessages")(function* (requ for (const part of message.content) { if (!ProviderShared.supportsContent(part, ["tool-result"])) return yield* ProviderShared.unsupportedContent("OpenAI Responses", "tool", ["tool-result"]) - input.push({ - type: "function_call_output", - call_id: part.id, - output: yield* lowerToolResultOutput(part), - }) + input.push( + customToolCallIDs.has(part.id) || + (!functionToolCallIDs.has(part.id) && + (historicalToolType(part) === "custom" || + (historicalToolType(part) === undefined && customTools.has(part.name)))) + ? { + type: "custom_tool_call_output", + call_id: part.id, + output: yield* lowerToolResultOutput(part), + } + : { + type: "function_call_output", + call_id: part.id, + output: yield* lowerToolResultOutput(part), + }, + ) } } @@ -472,7 +554,7 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request: model: request.model.id, input: yield* lowerMessages(request), tools: request.tools.length === 0 ? undefined : request.tools.map(lowerTool), - tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined, + tool_choice: request.toolChoice ? yield* lowerToolChoice(request.toolChoice, request.tools) : undefined, stream: true as const, max_output_tokens: generation?.maxTokens, temperature: generation?.temperature, @@ -506,6 +588,7 @@ const mapUsage = (usage: OpenAIResponsesUsage | null | undefined) => { const mapFinishReason = (event: OpenAIResponsesEvent, hasFunctionCall: boolean): FinishReason => { const reason = event.response?.incomplete_details?.reason + if (event.type === "response.incomplete" && (reason === undefined || reason === null)) return "unknown" if (reason === undefined || reason === null) return hasFunctionCall ? "tool-calls" : "stop" if (reason === "max_output_tokens") return "length" if (reason === "content_filter") return "content-filter" @@ -514,6 +597,9 @@ const mapFinishReason = (event: OpenAIResponsesEvent, hasFunctionCall: boolean): const openaiMetadata = (metadata: Record): ProviderMetadata => ({ openai: metadata }) +const toolMetadata = (itemId: string, type: "custom" | "function"): ProviderMetadata => + type === "custom" ? { openai: { itemId }, deepagent: { toolType: type } } : { openai: { itemId } } + // Hosted tool items (provider-executed) ship their typed input + status + // result fields all in one item. We expose them as a `tool-call` + // `tool-result` pair so consumers can treat them uniformly with client tools, @@ -653,8 +739,13 @@ const onOutputItemAdded = (state: ParserState, event: OpenAIResponsesEvent): Ste events, ] } - if (item?.type !== "function_call" || !item.id) return [state, NO_EVENTS] - const providerMetadata = openaiMetadata({ itemId: item.id }) + if ((item?.type !== "function_call" && item?.type !== "custom_tool_call") || !item.id) return [state, NO_EVENTS] + const providerMetadata = + item.type === "custom_tool_call" + ? toolMetadata(item.id, "custom") + : item.name === "apply_patch" + ? toolMetadata(item.id, "function") + : openaiMetadata({ itemId: item.id }) const events: LLMEvent[] = [] const lifecycle = Lifecycle.stepStart(state.lifecycle, events) return [ @@ -665,7 +756,8 @@ const onOutputItemAdded = (state: ParserState, event: OpenAIResponsesEvent): Ste tools: ToolStream.start(state.tools, item.id, { id: item.call_id ?? item.id, name: item.name ?? "", - input: item.arguments ?? "", + input: item.type === "custom_tool_call" ? (item.input ?? "") : (item.arguments ?? ""), + inputType: item.type === "custom_tool_call" ? "text" : "json", providerMetadata, }), }, @@ -789,6 +881,25 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenAIResponses.onFunctionCallAr return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult }) +const onCustomToolCallInputDelta = Effect.fn("OpenAIResponses.onCustomToolCallInputDelta")(function* ( + state: ParserState, + event: OpenAIResponsesEvent, +) { + if (!event.item_id || !event.delta) return [state, NO_EVENTS] satisfies StepResult + const result = ToolStream.appendExisting( + ADAPTER, + state.tools, + event.item_id, + event.delta, + "OpenAI Responses custom tool input delta is missing its tool call", + ) + if (ToolStream.isError(result)) return yield* result + const events: LLMEvent[] = [] + const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle + events.push(...result.events) + return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult +}) + const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function* ( state: ParserState, event: OpenAIResponsesEvent, @@ -796,27 +907,30 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function* const item = event.item if (!item) return [state, NO_EVENTS] satisfies StepResult - if (item.type === "function_call") { + if (item.type === "function_call" || item.type === "custom_tool_call") { if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult const tools = state.tools[item.id] ? state.tools - : ToolStream.start(state.tools, item.id, { id: item.call_id, name: item.name }) + : ToolStream.start(state.tools, item.id, { + id: item.call_id, + name: item.name, + inputType: item.type === "custom_tool_call" ? "text" : "json", + providerMetadata: toolMetadata(item.id, item.type === "custom_tool_call" ? "custom" : "function"), + }) + const input = item.type === "custom_tool_call" ? item.input : item.arguments const result = - item.arguments === undefined + input === undefined ? yield* ToolStream.finish(ADAPTER, tools, item.id) - : yield* ToolStream.finishWithInput(ADAPTER, tools, item.id, item.arguments) - const events: LLMEvent[] = [] + : yield* ToolStream.finishWithInput(ADAPTER, tools, item.id, input) const resultEvents = result.events ?? [] - const lifecycle = resultEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle - events.push(...resultEvents) return [ { ...state, - lifecycle, hasFunctionCall: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasFunctionCall, + pendingToolEvents: [...state.pendingToolEvents, ...resultEvents], tools: result.tools, }, - events, + NO_EVENTS, ] satisfies StepResult } @@ -857,9 +971,11 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function* }) const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): StepResult => { - const events: LLMEvent[] = [] + const reason = mapFinishReason(event, state.hasFunctionCall) + const releaseTools = event.type === "response.completed" && reason !== "length" && reason !== "content-filter" + const events: LLMEvent[] = releaseTools ? [...state.pendingToolEvents] : [] const lifecycle = Lifecycle.finish(state.lifecycle, events, { - reason: mapFinishReason(event, state.hasFunctionCall), + reason: releaseTools && state.pendingToolEvents.some(LLMEvent.is.toolCall) ? "tool-calls" : reason, usage: mapUsage(event.response?.usage), providerMetadata: event.response?.id || event.response?.service_tier @@ -869,7 +985,7 @@ const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): Step }) : undefined, }) - return [{ ...state, lifecycle }, events] + return [{ ...state, lifecycle, pendingToolEvents: [] }, events] } // Build a single human-readable message from whatever the provider supplied. @@ -924,6 +1040,7 @@ const step = (state: ParserState, event: OpenAIResponsesEvent) => { return Effect.succeed(onReasoningSummaryPartDone(state, event)) if (event.type === "response.output_item.added") return Effect.succeed(onOutputItemAdded(state, event)) if (event.type === "response.function_call_arguments.delta") return onFunctionCallArgumentsDelta(state, event) + if (event.type === "response.custom_tool_call_input.delta") return onCustomToolCallInputDelta(state, event) if (event.type === "response.output_item.done") return onOutputItemDone(state, event) if (event.type === "response.completed" || event.type === "response.incomplete") return Effect.succeed(onResponseFinish(state, event)) @@ -951,6 +1068,7 @@ export const protocol = Protocol.make({ initial: (request) => ({ hasFunctionCall: false, tools: ToolStream.empty(), + pendingToolEvents: [], lifecycle: Lifecycle.initial(), reasoningItems: {}, store: OpenAIOptions.store(request), diff --git a/packages/llm/src/protocols/utils/tool-stream.ts b/packages/llm/src/protocols/utils/tool-stream.ts index 8e07a64b..fada2a2f 100644 --- a/packages/llm/src/protocols/utils/tool-stream.ts +++ b/packages/llm/src/protocols/utils/tool-stream.ts @@ -10,6 +10,7 @@ type StreamKey = string | number * so far, not the parsed object. */ export interface PendingTool extends ToolAccumulator { + readonly inputType?: "json" | "text" readonly providerExecuted?: boolean readonly providerMetadata?: ProviderMetadata } @@ -64,7 +65,10 @@ const inputDelta = (tool: PendingTool, text: string) => }) const toolCall = (route: string, tool: PendingTool, inputOverride?: string) => - parseToolInput(route, tool.name, inputOverride ?? tool.input).pipe( + (tool.inputType === "text" + ? Effect.succeed(inputOverride ?? tool.input) + : parseToolInput(route, tool.name, inputOverride ?? tool.input) + ).pipe( Effect.map( (input): ToolCall => LLMEvent.toolCall({ diff --git a/packages/llm/src/schema/messages.ts b/packages/llm/src/schema/messages.ts index 8a0a91b5..7c94730f 100644 --- a/packages/llm/src/schema/messages.ts +++ b/packages/llm/src/schema/messages.ts @@ -313,21 +313,62 @@ export namespace Message { make({ role: "tool", content: ["type" in result ? result : ToolResultPart.make(result)] }) } -export class ToolDefinition extends Schema.Class("LLM.ToolDefinition")({ +export const CustomToolFormat = Schema.Union([ + Schema.Struct({ + type: Schema.Literal("grammar"), + syntax: Schema.Literals(["regex", "lark"]), + definition: Schema.String, + }), + Schema.Struct({ type: Schema.Literal("text") }), +]) +export type CustomToolFormat = Schema.Schema.Type + +const ToolDefinitionFields = { name: Schema.String, description: Schema.String, - inputSchema: JsonSchema, outputSchema: Schema.optional(JsonSchema), cache: Schema.optional(CacheHint), metadata: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)), native: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)), +} + +export class FunctionToolDefinition extends Schema.Class("LLM.FunctionToolDefinition")({ + ...ToolDefinitionFields, + type: Schema.Literal("function"), + inputSchema: JsonSchema, +}) {} + +export class CustomToolDefinition extends Schema.Class("LLM.CustomToolDefinition")({ + ...ToolDefinitionFields, + type: Schema.Literal("custom"), + format: CustomToolFormat, }) {} +export const ToolDefinitionSchema = Schema.Union([FunctionToolDefinition, CustomToolDefinition]) +export type ToolDefinition = FunctionToolDefinition | CustomToolDefinition + export namespace ToolDefinition { - export type Input = ToolDefinition | ConstructorParameters[0] + export type FunctionInput = Omit[0], "type"> & { + readonly type?: "function" + } + export type CustomInput = ConstructorParameters[0] + export type Input = ToolDefinition | FunctionInput | CustomInput /** Normalize tool definition input into the canonical `ToolDefinition` class. */ - export const make = (input: Input) => (input instanceof ToolDefinition ? input : new ToolDefinition(input)) + export const make = (input: Input): ToolDefinition => { + if (input instanceof FunctionToolDefinition || input instanceof CustomToolDefinition) return input + if (input.type === "custom") return new CustomToolDefinition(input) + return new FunctionToolDefinition({ ...input, type: "function" }) + } + + export const custom = (input: Omit) => new CustomToolDefinition({ ...input, type: "custom" }) + + export const is = (input: unknown): input is ToolDefinition => + input instanceof FunctionToolDefinition || input instanceof CustomToolDefinition + + export const isCustom = (tool: ToolDefinition): tool is CustomToolDefinition => tool.type === "custom" + + export const isFunction = (tool: ToolDefinition): tool is FunctionToolDefinition => tool.type === "function" } export class ToolChoice extends Schema.Class("LLM.ToolChoice")({ @@ -347,7 +388,7 @@ export namespace ToolChoice { /** Normalize ergonomic tool-choice inputs into the canonical `ToolChoice` class. */ export const make = (input: Input) => { if (input instanceof ToolChoice) return input - if (input instanceof ToolDefinition) return named(input.name) + if (ToolDefinition.is(input)) return named(input.name) if (typeof input === "string") return isMode(input) ? new ToolChoice({ type: input }) : named(input) return new ToolChoice(input) } @@ -356,7 +397,7 @@ export namespace ToolChoice { export const ResponseFormat = Schema.Union([ Schema.Struct({ type: Schema.Literal("text") }), Schema.Struct({ type: Schema.Literal("json"), schema: JsonSchema }), - Schema.Struct({ type: Schema.Literal("tool"), tool: ToolDefinition }), + Schema.Struct({ type: Schema.Literal("tool"), tool: ToolDefinitionSchema }), ]).pipe(Schema.toTaggedUnion("type")) export type ResponseFormat = Schema.Schema.Type @@ -365,7 +406,7 @@ export class LLMRequest extends Schema.Class("LLM.Request")({ model: ModelSchema, system: Schema.Array(SystemPart), messages: Schema.Array(Message), - tools: Schema.Array(ToolDefinition), + tools: Schema.Array(ToolDefinitionSchema), toolChoice: Schema.optional(ToolChoice), generation: Schema.optional(GenerationOptions), providerOptions: Schema.optional(ProviderOptions), diff --git a/packages/llm/src/tool.ts b/packages/llm/src/tool.ts index cebb293b..340f8916 100644 --- a/packages/llm/src/tool.ts +++ b/packages/llm/src/tool.ts @@ -1,5 +1,6 @@ import { Effect, JsonSchema, Schema } from "effect" import type { + CustomToolFormat, ToolCallPart, ToolContent, ToolDefinition as ToolDefinitionClass, @@ -93,6 +94,7 @@ type TypedToolConfig = { type DynamicToolConfig = { readonly description: string readonly jsonSchema: JsonSchema.JsonSchema + readonly format?: CustomToolFormat readonly outputSchema?: JsonSchema.JsonSchema readonly execute?: (params: unknown, context?: ToolExecuteContext) => Effect.Effect readonly toModelOutput?: (input: ToolModelOutputInput) => ReadonlyArray @@ -149,6 +151,7 @@ export function make, Success extends ToolSch export function make(config: { readonly description: string readonly jsonSchema: JsonSchema.JsonSchema + readonly format?: CustomToolFormat readonly outputSchema?: JsonSchema.JsonSchema readonly execute: (params: unknown, context?: ToolExecuteContext) => Effect.Effect readonly toModelOutput?: (input: ToolModelOutputInput) => ReadonlyArray @@ -157,6 +160,7 @@ export function make(config: { export function make(config: { readonly description: string readonly jsonSchema: JsonSchema.JsonSchema + readonly format?: CustomToolFormat readonly outputSchema?: JsonSchema.JsonSchema readonly execute?: undefined readonly toModelOutput?: (input: ToolModelOutputInput) => ReadonlyArray @@ -176,12 +180,19 @@ export function make(config: TypedToolConfig | DynamicToolConfig): AnyTool { _project: (parameters, callID, output) => project(config.toModelOutput, config.toStructuredOutput, parameters, callID, output), _legacyResult: config.toModelOutput === undefined && config.toStructuredOutput === undefined, - _definition: new ToolDefinition({ - name: "", - description: config.description, - inputSchema: config.jsonSchema, - outputSchema: config.outputSchema, - }), + _definition: config.format + ? ToolDefinition.custom({ + name: "", + description: config.description, + format: config.format, + outputSchema: config.outputSchema, + }) + : ToolDefinition.make({ + name: "", + description: config.description, + inputSchema: config.jsonSchema, + outputSchema: config.outputSchema, + }), } } return { @@ -196,7 +207,7 @@ export function make(config: TypedToolConfig | DynamicToolConfig): AnyTool { _project: (parameters, callID, output) => project(config.toModelOutput, config.toStructuredOutput, parameters, callID, output), _legacyResult: false, - _definition: new ToolDefinition({ + _definition: ToolDefinition.make({ name: "", description: config.description, inputSchema: toJsonSchema(config.parameters), @@ -219,14 +230,20 @@ export type Tools = Record * is reused. */ export const toDefinitions = (tools: Tools): ReadonlyArray => - Object.entries(tools).map( - ([name, item]) => - new ToolDefinition({ - name, - description: item._definition.description, - inputSchema: item._definition.inputSchema, - outputSchema: item._definition.outputSchema, - }), + Object.entries(tools).map(([name, item]) => + ToolDefinition.isCustom(item._definition) + ? ToolDefinition.custom({ + name, + description: item._definition.description, + format: item._definition.format, + outputSchema: item._definition.outputSchema, + }) + : ToolDefinition.make({ + name, + description: item._definition.description, + inputSchema: item._definition.inputSchema, + outputSchema: item._definition.outputSchema, + }), ) const toJsonSchema = (schema: Schema.Top): JsonSchema.JsonSchema => { diff --git a/packages/llm/test/generate-object.test.ts b/packages/llm/test/generate-object.test.ts index 9606f58f..7bfb57d1 100644 --- a/packages/llm/test/generate-object.test.ts +++ b/packages/llm/test/generate-object.test.ts @@ -1,6 +1,6 @@ import { describe, expect, test } from "bun:test" import { Effect, Schema } from "effect" -import { LLM } from "../src" +import { LLM, ToolDefinition } from "../src" import * as OpenAIChat from "../src/protocols/openai-chat" import { Auth } from "../src/route" import { Tool, toDefinitions } from "../src/tool" @@ -39,6 +39,7 @@ describe("Tool.make (dynamic JSON Schema)", () => { execute: () => Effect.succeed({ ok: true }), }) const [definition] = toDefinitions({ lookup }) + if (!definition || !ToolDefinition.isFunction(definition)) throw new Error("lookup must be a function tool") expect(definition?.name).toBe("lookup") expect(definition?.description).toBe("Look up something") expect(definition?.inputSchema).toEqual(jsonSchema) diff --git a/packages/llm/test/llm.test.ts b/packages/llm/test/llm.test.ts index 633a4662..e6ebbee7 100644 --- a/packages/llm/test/llm.test.ts +++ b/packages/llm/test/llm.test.ts @@ -2,7 +2,16 @@ import { describe, expect, test } from "bun:test" import { CacheHint, LLM, LLMResponse } from "../src" import * as OpenAIChat from "../src/protocols/openai-chat" import * as OpenAIResponses from "../src/protocols/openai-responses" -import { LLMRequest, Message, Model, ToolCallPart, ToolChoice, ToolDefinition, ToolResultPart } from "../src/schema" +import { + FunctionToolDefinition, + LLMRequest, + Message, + Model, + ToolCallPart, + ToolChoice, + ToolDefinition, + ToolResultPart, +} from "../src/schema" const chatRoute = OpenAIChat.route const responsesRoute = OpenAIResponses.route @@ -102,7 +111,7 @@ describe("llm constructors", () => { test("builds tool choices from names and tools", () => { const tool = ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } }) - expect(tool).toBeInstanceOf(ToolDefinition) + expect(tool).toBeInstanceOf(FunctionToolDefinition) expect(ToolChoice.make("lookup")).toEqual(new ToolChoice({ type: "tool", name: "lookup" })) expect(ToolChoice.named("required")).toEqual(new ToolChoice({ type: "tool", name: "required" })) expect(ToolChoice.make(tool)).toEqual(new ToolChoice({ type: "tool", name: "lookup" })) diff --git a/packages/llm/test/provider/anthropic-messages.test.ts b/packages/llm/test/provider/anthropic-messages.test.ts index f9f4650c..a721c96d 100644 --- a/packages/llm/test/provider/anthropic-messages.test.ts +++ b/packages/llm/test/provider/anthropic-messages.test.ts @@ -1,7 +1,7 @@ import { describe, expect } from "bun:test" import { Effect } from "effect" import { HttpClientRequest } from "effect/unstable/http" -import { CacheHint, LLM, LLMError, Message, ToolCallPart, Usage } from "../../src" +import { CacheHint, LLM, LLMError, Message, ToolCallPart, ToolDefinition, Usage } from "../../src" import { Auth, LLMClient } from "../../src/route" import * as AnthropicMessages from "../../src/protocols/anthropic-messages" import { continuationRequest, nativeAnthropicMessagesContinuation } from "../continuation-scenarios" @@ -218,6 +218,24 @@ describe("Anthropic Messages route", () => { }), ) + it.effect("rejects raw custom tools before sending Anthropic requests", () => + Effect.gen(function* () { + const error = yield* LLMClient.prepare( + LLM.updateRequest(request, { + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch", + format: { type: "text" }, + }), + ], + }), + ).pipe(Effect.flip) + + expect(error.message).toContain("Anthropic Messages does not support custom text tools") + }), + ) + // Regression: screenshot/read tool results must stay structured so base64 // image data is not JSON-stringified into `tool_result.content`. it.effect("lowers image tool-result content as structured image blocks", () => @@ -463,6 +481,24 @@ describe("Anthropic Messages route", () => { }), ) + it.effect("does not emit a tool call when an Anthropic input block is truncated", () => + Effect.gen(function* () { + const body = sseEvents( + { type: "message_start", message: { usage: { input_tokens: 5 } } }, + { type: "content_block_start", index: 0, content_block: { type: "tool_use", id: "call_1", name: "write" } }, + { type: "content_block_delta", index: 0, delta: { type: "input_json_delta", partial_json: '{"filePath":"x"' } }, + ) + const response = yield* LLMClient.generate( + LLM.updateRequest(request, { + tools: [{ name: "write", description: "Write file", inputSchema: { type: "object" } }], + }), + ).pipe(Effect.provide(fixedResponse(body))) + + expect(response.toolCalls).toEqual([]) + expect(response.events.some((event) => event.type === "tool-call")).toBe(false) + }), + ) + it.effect("emits provider-error events for mid-stream provider errors", () => Effect.gen(function* () { const response = yield* LLMClient.generate(request).pipe( diff --git a/packages/llm/test/provider/openai-chat.test.ts b/packages/llm/test/provider/openai-chat.test.ts index a2e9dd8d..2fbe0073 100644 --- a/packages/llm/test/provider/openai-chat.test.ts +++ b/packages/llm/test/provider/openai-chat.test.ts @@ -1,7 +1,7 @@ import { describe, expect } from "bun:test" import { Effect, Schema, Stream } from "effect" import { HttpClientRequest } from "effect/unstable/http" -import { LLM, LLMError, Message, Model, ToolCallPart, Usage } from "../../src" +import { LLM, LLMError, Message, Model, ToolCallPart, ToolDefinition, Usage } from "../../src" import * as Azure from "../../src/providers/azure" import * as OpenAI from "../../src/providers/openai" import * as OpenAIChat from "../../src/protocols/openai-chat" @@ -226,6 +226,24 @@ describe("OpenAI Chat route", () => { }), ) + it.effect("rejects raw custom tools before sending Chat requests", () => + Effect.gen(function* () { + const error = yield* LLMClient.prepare( + LLM.updateRequest(request, { + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch", + format: { type: "text" }, + }), + ], + }), + ).pipe(Effect.flip) + + expect(error.message).toContain("OpenAI Chat does not support custom text tools") + }), + ) + it.effect("continues image tool results as vision input without base64 text", () => Effect.gen(function* () { const prepared = yield* LLMClient.prepare( @@ -582,6 +600,27 @@ describe("OpenAI Chat route", () => { }), ) + it.effect("does not finalize streamed tool calls on a length finish", () => + Effect.gen(function* () { + const body = sseEvents( + deltaChunk({ + role: "assistant", + tool_calls: [{ index: 0, id: "call_1", function: { name: "write", arguments: '{"filePath":"x"' } }], + }), + deltaChunk({}, "length"), + ) + const response = yield* LLMClient.generate( + LLM.updateRequest(request, { + tools: [{ name: "write", description: "Write file", inputSchema: { type: "object" } }], + }), + ).pipe(Effect.provide(fixedResponse(body))) + + expect(response.toolCalls).toEqual([]) + expect(response.events.some((event) => event.type === "tool-call")).toBe(false) + expect(response.events.at(-1)).toMatchObject({ type: "finish", reason: "length" }) + }), + ) + it.effect("fails on malformed stream events", () => Effect.gen(function* () { const body = sseEvents(deltaChunk({ content: 123 })) diff --git a/packages/llm/test/provider/openai-responses.test.ts b/packages/llm/test/provider/openai-responses.test.ts index 5b66ba22..55c5e09f 100644 --- a/packages/llm/test/provider/openai-responses.test.ts +++ b/packages/llm/test/provider/openai-responses.test.ts @@ -1,7 +1,7 @@ import { describe, expect } from "bun:test" import { ConfigProvider, Effect, Layer, Stream } from "effect" import { Headers, HttpClientRequest } from "effect/unstable/http" -import { LLM, LLMError, Message, Model, ToolCallPart, Usage } from "../../src" +import { LLM, LLMError, Message, Model, ToolCallPart, ToolDefinition, Usage } from "../../src" import { Auth, LLMClient, RequestExecutor, WebSocketExecutor } from "../../src/route" import * as Azure from "../../src/providers/azure" import * as OpenAI from "../../src/providers/openai" @@ -130,6 +130,41 @@ describe("OpenAI Responses route", () => { }), ) + it.effect("prepares grammar-constrained custom tools without a JSON wrapper", () => + Effect.gen(function* () { + const prepared = yield* LLMClient.prepare( + LLM.updateRequest(request, { + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch.", + format: { + type: "grammar", + syntax: "lark", + definition: 'start: "*** Begin Patch"', + }, + }), + ], + toolChoice: "apply_patch", + }), + ) + + expect(prepared.body.tools).toEqual([ + { + type: "custom", + name: "apply_patch", + description: "Apply a patch.", + format: { + type: "grammar", + syntax: "lark", + definition: 'start: "*** Begin Patch"', + }, + }, + ]) + expect(prepared.body.tool_choice).toEqual({ type: "custom", name: "apply_patch" }) + }), + ) + it.effect("lowers chronological system updates to escaped user wrappers in order", () => Effect.gen(function* () { const prepared = yield* LLMClient.prepare( @@ -360,6 +395,97 @@ describe("OpenAI Responses route", () => { }), ) + it.effect("replays custom tool calls and outputs as native Responses items", () => + Effect.gen(function* () { + const patch = "*** Begin Patch\n*** Add File: hello.txt\n+hello\n*** End Patch" + const prepared = yield* LLMClient.prepare( + LLM.request({ + model, + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch.", + format: { type: "text" }, + }), + ], + messages: [ + Message.assistant([ + ToolCallPart.make({ id: "call_patch", name: "apply_patch", input: { patchText: patch } }), + ]), + Message.tool({ id: "call_patch", name: "apply_patch", result: "Done", resultType: "text" }), + ], + }), + ) + + expect(prepared.body.input).toEqual([ + { type: "custom_tool_call", call_id: "call_patch", name: "apply_patch", input: patch }, + { type: "custom_tool_call_output", call_id: "call_patch", output: "Done" }, + ]) + }), + ) + + it.effect("preserves the historical tool wire type when the current tool table changed", () => + Effect.gen(function* () { + const marker = (toolType: "custom" | "function") => ({ deepagent: { toolType } }) + const prepared = yield* LLMClient.prepare( + LLM.request({ + model, + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch.", + format: { type: "text" }, + }), + ], + messages: [ + Message.assistant([ + ToolCallPart.make({ + id: "call_custom", + name: "apply_patch", + input: { patchText: "*** Begin Patch\n*** End Patch" }, + providerMetadata: marker("custom"), + }), + ]), + Message.tool({ + id: "call_custom", + name: "apply_patch", + result: "Done", + resultType: "text", + providerMetadata: marker("custom"), + }), + Message.assistant([ + ToolCallPart.make({ + id: "call_function", + name: "apply_patch", + input: { patchText: "legacy" }, + providerMetadata: marker("function"), + }), + ]), + Message.tool({ + id: "call_function", + name: "apply_patch", + result: "Done", + resultType: "text", + providerMetadata: marker("function"), + }), + ], + }), + ) + + expect(prepared.body.input).toEqual([ + { + type: "custom_tool_call", + call_id: "call_custom", + name: "apply_patch", + input: "*** Begin Patch\n*** End Patch", + }, + { type: "custom_tool_call_output", call_id: "call_custom", output: "Done" }, + { type: "function_call", call_id: "call_function", name: "apply_patch", arguments: '{"patchText":"legacy"}' }, + { type: "function_call_output", call_id: "call_function", output: "Done" }, + ]) + }), + ) + // Regression: screenshot/read tool results must stay structured so base64 // image data is not JSON-stringified into `function_call_output.output`. it.effect("lowers image tool-result content as structured input_image items", () => @@ -1149,6 +1275,119 @@ describe("OpenAI Responses route", () => { }), ) + it.effect("assembles streamed custom tool input as raw text", () => + Effect.gen(function* () { + const patch = "*** Begin Patch\n*** Add File: hello.txt\n+hello\n*** End Patch" + const response = yield* LLMClient.generate( + LLM.updateRequest(request, { + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch.", + format: { type: "text" }, + }), + ], + }), + ).pipe( + Effect.provide( + fixedResponse( + sseEvents( + { + type: "response.output_item.added", + item: { + type: "custom_tool_call", + id: "item_patch", + call_id: "call_patch", + name: "apply_patch", + input: "", + }, + }, + { type: "response.custom_tool_call_input.delta", item_id: "item_patch", delta: patch.slice(0, 24) }, + { type: "response.custom_tool_call_input.delta", item_id: "item_patch", delta: patch.slice(24) }, + { + type: "response.output_item.done", + item: { + type: "custom_tool_call", + id: "item_patch", + call_id: "call_patch", + name: "apply_patch", + input: patch, + }, + }, + { type: "response.completed", response: {} }, + ), + ), + ), + ) + + expect(response.events.filter((event) => event.type.startsWith("tool-"))).toEqual([ + { + type: "tool-input-start", + id: "call_patch", + name: "apply_patch", + providerMetadata: { openai: { itemId: "item_patch" }, deepagent: { toolType: "custom" } }, + }, + { type: "tool-input-delta", id: "call_patch", name: "apply_patch", text: patch.slice(0, 24) }, + { type: "tool-input-delta", id: "call_patch", name: "apply_patch", text: patch.slice(24) }, + { + type: "tool-input-end", + id: "call_patch", + name: "apply_patch", + providerMetadata: { openai: { itemId: "item_patch" }, deepagent: { toolType: "custom" } }, + }, + { + type: "tool-call", + id: "call_patch", + name: "apply_patch", + input: patch, + providerExecuted: undefined, + providerMetadata: { openai: { itemId: "item_patch" }, deepagent: { toolType: "custom" } }, + }, + ]) + }), + ) + + it.effect("does not emit a custom tool call when the response is truncated", () => + Effect.gen(function* () { + const response = yield* LLMClient.generate( + LLM.updateRequest(request, { + tools: [ + ToolDefinition.custom({ + name: "apply_patch", + description: "Apply a patch.", + format: { type: "text" }, + }), + ], + }), + ).pipe( + Effect.provide( + fixedResponse( + sseEvents( + { + type: "response.output_item.added", + item: { + type: "custom_tool_call", + id: "item_patch", + call_id: "call_patch", + name: "apply_patch", + input: "", + }, + }, + { type: "response.custom_tool_call_input.delta", item_id: "item_patch", delta: "*** Begin Patch\n" }, + { + type: "response.incomplete", + response: { incomplete_details: { reason: "max_output_tokens" } }, + }, + ), + ), + ), + ) + + expect(response.events.some((event) => event.type === "tool-call")).toBe(false) + expect(response.events).toContainEqual(expect.objectContaining({ type: "finish", reason: "length" })) + }), + ) + it.effect("decodes web_search_call as provider-executed tool-call + tool-result", () => Effect.gen(function* () { const item = { From 90253746ba46ed30bcb40fc582983778f4791080 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:04:04 +0800 Subject: [PATCH 16/22] fix(deepagent-code): harden subagent execution and supervision --- .../e2e/regression/subagents-panel.spec.ts | 191 +++ packages/app/e2e/utils/mock-server.ts | 15 +- packages/app/playwright.config.ts | 4 +- packages/app/src/context/layout-helpers.ts | 51 +- packages/app/src/context/layout.test.ts | 49 +- packages/app/src/context/layout.tsx | 53 +- packages/app/src/i18n/en.ts | 39 +- packages/app/src/i18n/zh.ts | 27 +- packages/app/src/i18n/zht.ts | 18 +- .../pages/session/panel-error-boundary.tsx | 16 + .../src/pages/session/session-side-panel.tsx | 517 ++++--- .../side-panel-subagents-structure.test.ts | 56 + .../pages/session/side-panel-subagents.tsx | 85 +- .../src/pages/session/subagent-state.test.ts | 8 + .../app/src/pages/session/subagent-state.ts | 25 +- packages/core/src/database/migration.gen.ts | 1 + .../20260724134000_task_run_delivery.ts | 92 ++ packages/core/src/session/sql.ts | 98 ++ packages/core/src/v1/session.ts | 14 + packages/core/test/database-migration.test.ts | 14 + packages/deepagent-code/package.json | 1 + packages/deepagent-code/src/agent/agent.ts | 14 +- .../src/effect/instance-registry.ts | 20 + .../src/effect/runtime-flags.ts | 4 + .../src/project/instance-store.ts | 9 +- .../deepagent-code/src/session/message-v2.ts | 29 +- .../deepagent-code/src/session/processor.ts | 153 +- packages/deepagent-code/src/session/prompt.ts | 478 ++++-- .../deepagent-code/src/session/session.ts | 13 +- packages/deepagent-code/src/session/tools.ts | 15 +- .../src/tool/semantic-fingerprint.ts | 26 + packages/deepagent-code/src/tool/shell.ts | 14 + packages/deepagent-code/src/tool/task-run.ts | 897 +++++++++++ packages/deepagent-code/src/tool/task.ts | 1374 +++++++++++++---- packages/deepagent-code/src/tool/tool.ts | 2 + .../test/project/instance-bootstrap.test.ts | 54 +- .../test/session/message-v2.test.ts | 49 + .../test/session/processor-effect.test.ts | 3 +- .../test/session/prompt.test.ts | 159 ++ .../session/tool-sequence-tracker.test.ts | 49 + .../test/tool/task-finalizer.test.ts | 231 +++ .../deepagent-code/test/tool/task-run.test.ts | 489 ++++++ .../test/tool/task-takeover.test.ts | 92 +- .../deepagent-code/test/tool/task.test.ts | 62 +- packages/desktop/package.json | 4 + packages/desktop/scripts/subagents-smoke.ts | 189 +++ .../scripts/verify-subagents-sourcemap.ts | 53 + packages/desktop/src/main/index.ts | 10 +- 48 files changed, 5012 insertions(+), 854 deletions(-) create mode 100644 packages/app/e2e/regression/subagents-panel.spec.ts create mode 100644 packages/app/src/pages/session/panel-error-boundary.tsx create mode 100644 packages/app/src/pages/session/side-panel-subagents-structure.test.ts create mode 100644 packages/core/src/database/migration/20260724134000_task_run_delivery.ts create mode 100644 packages/deepagent-code/src/tool/semantic-fingerprint.ts create mode 100644 packages/deepagent-code/src/tool/task-run.ts create mode 100644 packages/deepagent-code/test/tool/task-finalizer.test.ts create mode 100644 packages/deepagent-code/test/tool/task-run.test.ts create mode 100644 packages/desktop/scripts/subagents-smoke.ts create mode 100644 packages/desktop/scripts/verify-subagents-sourcemap.ts diff --git a/packages/app/e2e/regression/subagents-panel.spec.ts b/packages/app/e2e/regression/subagents-panel.spec.ts new file mode 100644 index 00000000..59259158 --- /dev/null +++ b/packages/app/e2e/regression/subagents-panel.spec.ts @@ -0,0 +1,191 @@ +import { expect, test, type Page } from "@playwright/test" +import { base64Encode } from "@deepagent-code/core/util/encode" +import { mockDeepAgentCodeServer } from "../utils/mock-server" +import { expectAppVisible } from "../utils/waits" + +const directory = "C:/DeepAgent Code/SubagentPanel" +const projectID = "proj_subagent_panel" +const parentID = "ses_subagent_parent" +const slug = base64Encode(directory) +const sessionKey = `local\u0000${slug}/${parentID}` + +const parent = { + id: parentID, + slug: "subagent-parent", + projectID, + directory, + title: "Subagent parent", + version: "dev", + time: { created: 1_700_000_000_000, updated: 1_700_000_000_000 }, +} + +const child = (id: string, title: string, updated: number, state: string, reason: string) => ({ + id, + slug: id, + projectID, + directory, + parentID, + title, + version: "dev", + metadata: { + deepagent: { + subagent: { finished: state !== "researching", state, reason, run_id: `run_${id}`, generation: 1 }, + }, + }, + time: { created: updated - 1, updated }, +}) + +async function openPersistedPanel(page: Page, children: ReturnType[], mode = "subagents") { + const pageErrors: string[] = [] + const consoleErrors: string[] = [] + page.on("pageerror", (error) => pageErrors.push(error.stack ?? error.message)) + page.on("console", (message) => { + if (message.type() === "error") consoleErrors.push(message.text()) + }) + await mockDeepAgentCodeServer(page, { + directory, + project: { + id: projectID, + worktree: directory, + vcs: "git", + name: "subagent-panel", + time: { created: 1_700_000_000_000, updated: 1_700_000_000_000 }, + sandboxes: [], + }, + provider: { + all: [ + { + id: "deepagent-code", + name: "DeepAgent Code", + models: { model: { id: "model", name: "Model", limit: { context: 200_000 } } }, + }, + ], + connected: ["deepagent-code"], + default: { providerID: "deepagent-code", modelID: "model" }, + }, + sessions: [parent, ...children], + pageMessages: () => ({ items: [] }), + }) + await page.addInitScript( + ({ key, value }) => { + localStorage.setItem("settings.v3", JSON.stringify({ general: { newLayoutDesigns: true } })) + localStorage.setItem( + "deepagent.global.dat:layout", + JSON.stringify({ sessionView: { [key]: { scroll: {}, rightPanelMode: value } } }), + ) + }, + { key: sessionKey, value: mode }, + ) + await page.goto(`/${slug}/session/${parentID}`) + await expectAppVisible(page.getByRole("tab", { name: "Subagents" })) + return { pageErrors, consoleErrors } +} + +test("persisted subagent panel renders many terminal states and keeps row actions independent", async ({ page }) => { + const children = [ + child("ses_child_error", "Failed researcher", 104, "error", "provider_error"), + child("ses_child_cancelled", "Cancelled reviewer", 103, "cancelled", "human"), + child("ses_child_interrupted", "Interrupted researcher", 102, "interrupted", "doom_loop"), + child("ses_child_completed", "Completed reviewer", 101, "completed", "structured_output_valid"), + ] + const errors = await openPersistedPanel(page, children) + await expect(page.getByText("Failed researcher")).toBeVisible() + await expect(page.getByText("provider_error")).toBeVisible() + await expect(page.getByRole("button", { name: "Select subagent Cancelled reviewer" })).toContainText("cancelled") + await expect(page.getByRole("button", { name: "Select subagent Interrupted researcher" })).toContainText( + "interrupted", + ) + await expect(page.getByRole("button", { name: "Select subagent Completed reviewer" })).toContainText("finished") + + const completedRow = page.getByRole("button", { name: "Select subagent Completed reviewer" }) + const completedContainer = completedRow.locator("..") + await completedRow.focus() + await completedRow.press("Enter") + await expect(completedContainer).toHaveClass(/ring-border-strong-base/) + await expect(page).toHaveURL(new RegExp(`/${parentID}$`)) + await completedRow.press("Space") + await expect(completedContainer).not.toHaveClass(/ring-border-strong-base/) + + await page.getByRole("button", { name: "Open Completed reviewer" }).click() + await expect(page).toHaveURL(new RegExp("/ses_child_completed$")) + expect(errors).toEqual({ pageErrors: [], consoleErrors: [] }) +}) + +for (const count of [0, 1]) { + test(`persisted subagent panel cold-renders ${count} child without browser errors`, async ({ page }) => { + const errors = await openPersistedPanel( + page, + count === 0 ? [] : [child("ses_only_child", "Only researcher", 101, "completed", "text_output_valid")], + ) + if (count === 0) await expect(page.getByText("No subagents for this session")).toBeVisible() + if (count === 1) await expect(page.getByText("Only researcher")).toBeVisible() + expect(errors).toEqual({ pageErrors: [], consoleErrors: [] }) + }) +} + +test("unknown persisted panel mode fails closed", async ({ page }) => { + const errors = await openPersistedPanel(page, [], "removed-panel") + await expect(page.getByRole("tab", { name: "Subagents" })).toHaveAttribute("aria-selected", "false") + await expect(page.getByText("No subagents for this session")).toHaveCount(0) + expect(errors).toEqual({ pageErrors: [], consoleErrors: [] }) +}) + +test("production DOM contains no nested interactive controls in the subagent panel", async ({ page }) => { + await openPersistedPanel(page, [child("ses_dom_child", "DOM researcher", 101, "completed", "text_output_valid")]) + const nested = await page.locator("#review-panel").evaluate((panel) => { + const selector = "a[href],button,input,select,textarea,summary,[role=button],[role=link],[role=tab]" + return [...panel.querySelectorAll(selector)] + .filter((element) => element.parentElement?.closest(selector)) + .map((element) => element.outerHTML) + }) + expect(nested).toEqual([]) +}) + +test("panel render failure stays local and retry recovers without a renderer error", async ({ page }) => { + await page.addInitScript(() => { + const original = Element.prototype.setAttribute + Element.prototype.setAttribute = function (name, value) { + const root = window as typeof window & { __subagentPanelFailureInjected?: boolean } + if ( + !root.__subagentPanelFailureInjected && + name === "aria-label" && + String(value).startsWith("Select subagent") + ) { + root.__subagentPanelFailureInjected = true + throw new Error("injected subagent panel render failure") + } + return original.call(this, name, value) + } + }) + const errors = await openPersistedPanel(page, [ + child("ses_boundary_child", "Boundary researcher", 101, "completed", "text_output_valid"), + ]) + await expect(page.getByText("The subagent panel could not be displayed.")).toBeVisible() + await expect(page.getByRole("tab", { name: "Subagents" })).toBeVisible() + expect(errors.pageErrors).toEqual([]) + expect(errors.consoleErrors.some((message) => message.includes("ui.panel.render_failed"))).toBe(true) + + await page.getByRole("button", { name: "Retry" }).click() + await expect(page.getByText("Boundary researcher")).toBeVisible() + await expect(page.getByText("The subagent panel could not be displayed.")).toHaveCount(0) +}) + +test("panel render failure can be closed without restarting the application", async ({ page }) => { + await page.addInitScript(() => { + const original = Element.prototype.setAttribute + Element.prototype.setAttribute = function (name, value) { + if (name === "aria-label" && String(value).startsWith("Select subagent")) { + throw new Error("injected persistent subagent panel failure") + } + return original.call(this, name, value) + } + }) + const errors = await openPersistedPanel(page, [ + child("ses_close_child", "Close researcher", 101, "completed", "text_output_valid"), + ]) + await expect(page.getByText("The subagent panel could not be displayed.")).toBeVisible() + await page.getByRole("button", { name: "Close", exact: true }).last().click() + await expect(page.getByText("The subagent panel could not be displayed.")).toHaveCount(0) + await expect(page.getByRole("tab", { name: "Subagents" })).toHaveAttribute("aria-selected", "false") + expect(errors.pageErrors).toEqual([]) +}) diff --git a/packages/app/e2e/utils/mock-server.ts b/packages/app/e2e/utils/mock-server.ts index f0851056..5433c3dd 100644 --- a/packages/app/e2e/utils/mock-server.ts +++ b/packages/app/e2e/utils/mock-server.ts @@ -42,11 +42,22 @@ export async function mockDeepAgentCodeServer(page: Page, config: MockServerConf await page.route("**/*", async (route) => { const url = new URL(route.request().url()) - const targetPort = process.env.PLAYWRIGHT_SERVER_PORT ?? "4096" + const targetPort = + process.env.PLAYWRIGHT_SERVER_PORT ?? + (process.env.PLAYWRIGHT_PRODUCTION ? (process.env.PLAYWRIGHT_PORT ?? "3000") : "4096") if (url.port !== targetPort) return route.fallback() + if ( + route.request().resourceType() === "document" || + url.pathname.startsWith("/assets/") || + /\.(?:css|html|ico|js|json|map|png|svg|wasm|woff2?)$/.test(url.pathname) + ) { + return route.fallback() + } const path = url.pathname - if (path === "/global/event" || path === "/event") return sse(route, config.events?.()) + if (path === "/global/event" || path === "/event" || path === "/api/event" || path === "/debug/events") { + return sse(route, config.events?.()) + } if (path === "/global/health") return json(route, { healthy: true, version: "test", runtimeId: "runtime-test" }) if (path === "/pty" && route.request().method() === "GET") return json(route, [...ptys.values()]) if (path === "/pty" && route.request().method() === "POST") { diff --git a/packages/app/playwright.config.ts b/packages/app/playwright.config.ts index 74b07346..0d332b2d 100644 --- a/packages/app/playwright.config.ts +++ b/packages/app/playwright.config.ts @@ -4,7 +4,9 @@ const port = Number(process.env.PLAYWRIGHT_PORT ?? 3000) const baseURL = process.env.PLAYWRIGHT_BASE_URL ?? `http://127.0.0.1:${port}` const serverHost = process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1" const serverPort = process.env.PLAYWRIGHT_SERVER_PORT ?? "4096" -const command = `bun run dev -- --host 0.0.0.0 --port ${port}` +const command = process.env.PLAYWRIGHT_PRODUCTION + ? `bun run build && bun run serve -- --host 0.0.0.0 --port ${port}` + : `bun run dev -- --host 0.0.0.0 --port ${port}` const reuse = !process.env.CI const workers = Number(process.env.PLAYWRIGHT_WORKERS ?? (process.env.CI ? 5 : 0)) || undefined const reporter = [["html", { outputFolder: "e2e/playwright-report", open: "never" }], ["line"]] as const diff --git a/packages/app/src/context/layout-helpers.ts b/packages/app/src/context/layout-helpers.ts index cf29313a..a33143f5 100644 --- a/packages/app/src/context/layout-helpers.ts +++ b/packages/app/src/context/layout-helpers.ts @@ -5,6 +5,50 @@ export type PanelLocation = "bottom" | "side" export const PANEL_VIEWS: readonly PanelView[] = ["terminal", "debug-console", "problems"] +const RIGHT_PANEL_MODES = new Set([ + "review", + "files", + "worktree", + "subagents", + "browser", + "mcp", + "plugins", + "profile", + "debug", + "im", + "terminal", + "debug-console", + "problems", +]) +const RIGHT_PANEL_FAILURE_THRESHOLD = 2 + +const isRecord = (value: unknown): value is Record => + typeof value === "object" && value !== null && !Array.isArray(value) + +export function migrateRightPanelSessionView(sessionView: unknown, build: string) { + if (!isRecord(sessionView)) return sessionView + let changed = false + const next = Object.fromEntries( + Object.entries(sessionView).map(([key, raw]) => { + if (!isRecord(raw)) return [key, raw] + const storedMode = raw.rightPanelMode + const mode = storedMode === "oversight" ? "subagents" : storedMode + const validMode = typeof mode === "string" && RIGHT_PANEL_MODES.has(mode) ? mode : undefined + const failure = isRecord(raw.rightPanelFailure) ? raw.rightPanelFailure : undefined + const quarantined = + validMode !== undefined && + failure?.mode === validMode && + failure.build === build && + typeof failure.count === "number" && + failure.count >= RIGHT_PANEL_FAILURE_THRESHOLD + if (storedMode === validMode && !quarantined) return [key, raw] + changed = true + return [key, { ...raw, rightPanelMode: quarantined ? undefined : validMode }] + }), + ) + return changed ? next : sessionView +} + export type BottomPanelState = { opened: boolean activeView?: PanelView @@ -77,9 +121,7 @@ export const togglePanel = ( return state } -export const toggleBottomPanel = ( - input: PanelTransitionInput, -): PanelSessionState => { +export const toggleBottomPanel = (input: PanelTransitionInput): PanelSessionState => { const state = clonePanelState(input.state) if (state.bottomPanel.opened) { state.bottomPanel.opened = false @@ -90,7 +132,8 @@ export const toggleBottomPanel = ( state.bottomPanel = { opened: true, activeView: - state.bottomPanel.activeView && panelHost(input.locations[state.bottomPanel.activeView], input.sideAvailable) === "bottom" + state.bottomPanel.activeView && + panelHost(input.locations[state.bottomPanel.activeView], input.sideAvailable) === "bottom" ? state.bottomPanel.activeView : fallback, } diff --git a/packages/app/src/context/layout.test.ts b/packages/app/src/context/layout.test.ts index 14e0263a..94f31074 100644 --- a/packages/app/src/context/layout.test.ts +++ b/packages/app/src/context/layout.test.ts @@ -11,10 +11,43 @@ import { toggleBottomPanel, togglePanel, toggledPanelMode, + migrateRightPanelSessionView, type PanelSessionState, type PanelTransitionInput, } from "./layout-helpers" +describe("right panel persisted-state quarantine", () => { + test("migrates oversight and removes unknown modes", () => { + expect( + migrateRightPanelSessionView( + { + old: { scroll: {}, rightPanelMode: "oversight" }, + invalid: { scroll: {}, rightPanelMode: "removed-mode" }, + }, + "1.0.0", + ), + ).toEqual({ + old: { scroll: {}, rightPanelMode: "subagents" }, + invalid: { scroll: {}, rightPanelMode: undefined }, + }) + }) + + test("closes a panel after two failures in the same build", () => { + const failed = (count: number, build = "1.0.0") => ({ + session: { + scroll: {}, + rightPanelMode: "subagents", + rightPanelFailure: { mode: "subagents", build, count }, + }, + }) + expect(migrateRightPanelSessionView(failed(1), "1.0.0")).toEqual(failed(1)) + expect(migrateRightPanelSessionView(failed(2), "1.0.0")).toEqual({ + session: { ...failed(2).session, rightPanelMode: undefined }, + }) + expect(migrateRightPanelSessionView(failed(2, "0.9.0"), "1.0.0")).toEqual(failed(2, "0.9.0")) + }) +}) + describe("right-side-panel mode reducer", () => { // Regression guard for the IM panel "进去出不来" bug: opening a panel must be // reversible. These exercise the exact reducer the LayoutProvider's @@ -82,7 +115,12 @@ describe("movable panel state machine", () => { test("bottom toggle replaces a stale active view with a valid fallback", () => { const locationsWithTerminalSide = { terminal: "side", "debug-console": "bottom", problems: "bottom" } as const expect( - toggleBottomPanel(input({ locations: locationsWithTerminalSide, state: { bottomPanel: { opened: false, activeView: "terminal" } } })), + toggleBottomPanel( + input({ + locations: locationsWithTerminalSide, + state: { bottomPanel: { opened: false, activeView: "terminal" } }, + }), + ), ).toEqual({ bottomPanel: { opened: true, activeView: "debug-console" } }) }) @@ -90,7 +128,10 @@ describe("movable panel state machine", () => { const opened = revealPanel(input(), "terminal") const moved = movePanel(input({ state: opened }), "terminal", "side") expect(moved.locations.terminal).toBe("side") - expect(moved.state).toEqual({ bottomPanel: { opened: true, activeView: "debug-console" }, rightPanelMode: "terminal" }) + expect(moved.state).toEqual({ + bottomPanel: { opened: true, activeView: "debug-console" }, + rightPanelMode: "terminal", + }) }) test("moving the final bottom view closes the bottom panel", () => { @@ -118,7 +159,9 @@ describe("movable panel state machine", () => { test("mobile refuses a side move and reveals an old side preference in bottom", () => { const mobile = input({ sideAvailable: false }) expect(movePanel(mobile, "terminal", "side")).toEqual({ locations, state }) - expect(revealPanel(input({ sideAvailable: false, locations: { ...locations, terminal: "side" } }), "terminal")).toEqual({ + expect( + revealPanel(input({ sideAvailable: false, locations: { ...locations, terminal: "side" } }), "terminal"), + ).toEqual({ bottomPanel: { opened: true, activeView: "terminal" }, }) }) diff --git a/packages/app/src/context/layout.tsx b/packages/app/src/context/layout.tsx index 9c296052..d715514a 100644 --- a/packages/app/src/context/layout.tsx +++ b/packages/app/src/context/layout.tsx @@ -26,6 +26,7 @@ import { toggleBottomPanel, togglePanel, toggledPanelMode, + migrateRightPanelSessionView, type PanelLocation, type PanelSessionState, type PanelTransitionInput, @@ -123,6 +124,11 @@ type SessionView = { | "terminal" | "debug-console" | "problems" + rightPanelFailure?: { + mode: string + build: string + count: number + } bottomPanel?: { opened: boolean activeView?: DockPanelID @@ -132,6 +138,9 @@ type SessionView = { todoCollapsed?: boolean } +const isRecord = (value: unknown): value is Record => + typeof value === "object" && value !== null && !Array.isArray(value) + type TabHandoff = { scope: ServerScope dir: string @@ -220,9 +229,6 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( return { ...value, server: server.key } }) - const isRecord = (value: unknown): value is Record => - typeof value === "object" && value !== null && !Array.isArray(value) - const migrate = (value: unknown) => { if (!isRecord(value)) return value @@ -265,6 +271,7 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( const sessionTabs = migrateLegacySessionStateKeys(value.sessionTabs) const sessionView = migrateLegacySessionStateKeys(value.sessionView) + const migratedSessionView = migrateRightPanelSessionView(sessionView, platform.version ?? "dev") const migratedSessionTabs = (() => { if (!isRecord(sessionTabs)) return sessionTabs @@ -307,7 +314,7 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( migratedFileTree === fileTree && migratedSessionTabs === value.sessionTabs && migratedRightPanel === rightPanel && - sessionView === value.sessionView + migratedSessionView === value.sessionView ) { return value } @@ -319,7 +326,7 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( fileTree: migratedFileTree, sessionTabs: migratedSessionTabs, rightPanel: migratedRightPanel, - sessionView, + sessionView: migratedSessionView, } } @@ -760,7 +767,10 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( setStore("dock", "location", id, next) }, bottomCount: createMemo( - () => DOCK_PANEL_IDS.filter((id) => (id === "terminal" ? true : (store.dock?.location?.[id] ?? DOCK_DEFAULT_LOCATION[id]) === "bottom")).length, + () => + DOCK_PANEL_IDS.filter((id) => + id === "terminal" ? true : (store.dock?.location?.[id] ?? DOCK_DEFAULT_LOCATION[id]) === "bottom", + ).length, ), }, review: { @@ -828,8 +838,8 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( width: (bucket: RightPanelWidthBucket = "wide") => { const ratio = bucket === "narrow" - ? store.rightPanel?.narrowRatio ?? DEFAULT_RIGHT_PANEL_NARROW_RATIO - : store.rightPanel?.ratio ?? DEFAULT_RIGHT_PANEL_RATIO + ? (store.rightPanel?.narrowRatio ?? DEFAULT_RIGHT_PANEL_NARROW_RATIO) + : (store.rightPanel?.ratio ?? DEFAULT_RIGHT_PANEL_RATIO) const max = Math.max(MIN_RIGHT_PANEL_PX, Math.round(windowWidth() * MAX_RIGHT_PANEL_RATIO)) return Math.min(max, Math.max(MIN_RIGHT_PANEL_PX, Math.round(windowWidth() * ratio))) }, @@ -1091,6 +1101,33 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext( toggle(mode: NonNullable) { setRightPanelMode(toggledPanelMode(rightPanelMode(), mode)) }, + reportFailure(mode: string, build: string) { + const session = key() + const current = store.sessionView[session] + const previous = current?.rightPanelFailure + const count = previous?.mode === mode && previous.build === build ? previous.count + 1 : 1 + if (!current) { + setStore("sessionView", session, { + scroll: {}, + rightPanelFailure: { mode, build, count }, + }) + return + } + setStore("sessionView", session, "rightPanelFailure", { mode, build, count }) + }, + clearFailure(mode: string, build: string) { + const session = key() + const current = store.sessionView[session] + if (!current?.rightPanelFailure) return + if (current.rightPanelFailure.mode !== mode || current.rightPanelFailure.build !== build) return + setStore( + "sessionView", + session, + produce((draft) => { + delete draft.rightPanelFailure + }), + ) + }, }, review: { open: createMemo(() => s().reviewOpen ?? []), diff --git a/packages/app/src/i18n/en.ts b/packages/app/src/i18n/en.ts index 07bdc19e..0f1bd014 100644 --- a/packages/app/src/i18n/en.ts +++ b/packages/app/src/i18n/en.ts @@ -191,7 +191,8 @@ export const dict = { "provider.connect.transport.error.number": "Enter a positive whole number, or leave blank.", "provider.custom.title": "Custom provider", - "provider.custom.description.prefix": "Enter the base URL and API key. We detect the protocol and available models automatically. See the ", + "provider.custom.description.prefix": + "Enter the base URL and API key. We detect the protocol and available models automatically. See the ", "provider.custom.description.link": "provider config docs", "provider.custom.description.suffix": ".", "provider.custom.advanced.toggle": "Advanced settings", @@ -208,7 +209,8 @@ export const dict = { "provider.custom.field.apiKey.placeholder": "API key", "provider.custom.field.apiKey.description": "Optional. Leave empty if you manage auth via headers.", "provider.custom.field.protocol.label": "Protocol", - "provider.custom.field.protocol.description": "Auto-detects by default. Pick a protocol if detection is wrong or the endpoint has no model list.", + "provider.custom.field.protocol.description": + "Auto-detects by default. Pick a protocol if detection is wrong or the endpoint has no model list.", "provider.custom.field.protocol.option.auto": "Auto-detect", "provider.custom.field.protocol.option.openai-compatible": "OpenAI-compatible", "provider.custom.field.protocol.option.anthropic": "Anthropic", @@ -885,14 +887,16 @@ export const dict = { "oversight.trace.source": "source:", "oversight.trace.causedBy": "caused by:", "oversight.takeover.title": "Human Takeover", - "oversight.takeover.description": "Record that a human is taking over from the autonomous agents (pauses autonomy escalation).", + "oversight.takeover.description": + "Record that a human is taking over from the autonomous agents (pauses autonomy escalation).", "oversight.takeover.placeholder": "Reason for taking over…", "oversight.takeover.action": "Take over", "oversight.takeover.recorded": "Takeover recorded.", "oversight.takeover.unsupported": "Takeover recording activates once the backend endpoint (P3.10) is available.", "oversight.takeover.failed": "Failed: {{error}}", "oversight.rollback.title": "Rollback", - "oversight.rollback.description": "Revert an agent-produced change over a session (via SessionRevert). Only affects a session in this workspace. This reverts real changes — use with care.", + "oversight.rollback.description": + "Revert an agent-produced change over a session (via SessionRevert). Only affects a session in this workspace. This reverts real changes — use with care.", "oversight.rollback.sessionPlaceholder": "Session ID (ses_…)", "oversight.rollback.reasonPlaceholder": "Reason for rolling back… (optional)", "oversight.rollback.action": "Roll back", @@ -916,6 +920,13 @@ export const dict = { "session.subagents.running": "running", "session.subagents.idle": "idle", "session.subagents.finished": "finished", + "session.subagents.error": "failed", + "session.subagents.interrupted": "interrupted", + "session.subagents.cancelled": "cancelled", + "session.subagents.select": "Select subagent", + "session.subagents.open": "Open", + "session.subagents.panelError": "The subagent panel could not be displayed.", + "session.subagents.retry": "Retry panel", "session.subagents.readonly": "This subagent has finished. Its history is read-only.", "session.fork.derivedFrom": "Derived from {title}", "session.fork.derivedFromUnknown": "Derived from another session", @@ -1084,7 +1095,8 @@ export const dict = { "goal.start.success": "Goal started — running in the background", "goal.start.failed": "Couldn't start the goal", "wiki.title": "Repo & Wiki", - "wiki.description": "Read, search, and govern the four graphs. Knowledge and Memory are editable; Documents and Code are read-only.", + "wiki.description": + "Read, search, and govern the four graphs. Knowledge and Memory are editable; Documents and Code are read-only.", "wiki.search": "Search pages and symbols", "wiki.searchEmpty": "No pages match your search", "wiki.empty": "No pages yet", @@ -1094,16 +1106,17 @@ export const dict = { "wiki.version": "v{{version}}", "wiki.links": "Linked code & docs", "wiki.stale": "stale", - "wiki.type.knowledge": "Knowledge", - "wiki.type.memory": "Memory", - "wiki.type.document": "Documents", - "wiki.type.code": "Code", + "wiki.type.knowledge": "Knowledge graph", + "wiki.type.memory": "Memory graph", + "wiki.type.document": "Document graph", + "wiki.type.code": "Code graph", "wiki.edit.button": "Edit", "wiki.edit.save": "Save", "wiki.edit.saved": "Knowledge page updated", "wiki.edit.failed": "Couldn't save the edit", "sidebar.history": "History projects", - "sidebar.history.description": "Projects with imported or past sessions that are not currently active. Click one to open it.", + "sidebar.history.description": + "Projects with imported or past sessions that are not currently active. Click one to open it.", "sidebar.workspaces.enable": "Enable workspaces", "sidebar.workspaces.disable": "Disable workspaces", "sidebar.gettingStarted.title": "Getting started", @@ -1214,7 +1227,8 @@ export const dict = { "settings.general.deepagent.prompt.intelligence.description": "DeepAgent prepares a complete prompt first, then lets you review and edit it.", "settings.general.deepagent.intelligenceModel.title": "Intelligence mode model", - "settings.general.deepagent.intelligenceModel.description": "Choose the model used only to prepare intelligence-mode prompt drafts.", + "settings.general.deepagent.intelligenceModel.description": + "Choose the model used only to prepare intelligence-mode prompt drafts.", "settings.general.deepagent.selfLearning.title": "Self-learning approval", "settings.general.deepagent.selfLearning.description": "Decide how knowledge DeepAgent learns from your sessions becomes retrievable.", @@ -1278,7 +1292,8 @@ export const dict = { "review.envFacts.scope.global": "All projects", "review.envFacts.scope.project": "This project only", "review.envFacts.scope.globalHint": "Corrects the shared fact everywhere. Other projects see the update next time.", - "review.envFacts.scope.projectHint": "Saves a local copy for this project. The shared fact stays unchanged for others.", + "review.envFacts.scope.projectHint": + "Saves a local copy for this project. The shared fact stays unchanged for others.", "packs.title": "Domain Packs", "packs.active": "Auto-detected", "packs.builtin": "Built-in packs", diff --git a/packages/app/src/i18n/zh.ts b/packages/app/src/i18n/zh.ts index 93fad193..def41e6d 100644 --- a/packages/app/src/i18n/zh.ts +++ b/packages/app/src/i18n/zh.ts @@ -203,8 +203,7 @@ export const dict = { "provider.connect.toast.connected.description": "现在可以使用 {{provider}} 模型了。", "provider.connect.transport.title": "高级：请求超时与重试", - "provider.connect.transport.description": - "官方 provider 的传输设置独立于配置文件保存。留空则使用默认值。", + "provider.connect.transport.description": "官方 provider 的传输设置独立于配置文件保存。留空则使用默认值。", "provider.connect.transport.headerTimeout.label": "响应头超时（毫秒）", "provider.connect.transport.headerTimeout.placeholder": "例如 10000", "provider.connect.transport.chunkTimeout.label": "数据块超时（毫秒）", @@ -795,7 +794,8 @@ export const dict = { "oversight.takeover.unsupported": "接管记录将在后端端点（P3.10）可用后启用。", "oversight.takeover.failed": "失败：{{error}}", "oversight.rollback.title": "回滚", - "oversight.rollback.description": "回滚某个会话中 Agent 产生的更改（通过 SessionRevert）。仅影响本工作区内的会话。此操作会回滚真实更改——请谨慎使用。", + "oversight.rollback.description": + "回滚某个会话中 Agent 产生的更改（通过 SessionRevert）。仅影响本工作区内的会话。此操作会回滚真实更改——请谨慎使用。", "oversight.rollback.sessionPlaceholder": "会话 ID（ses_…）", "oversight.rollback.reasonPlaceholder": "回滚原因…（可选）", "oversight.rollback.action": "回滚", @@ -817,6 +817,13 @@ export const dict = { "session.subagents.running": "运行中", "session.subagents.idle": "空闲", "session.subagents.finished": "已完成", + "session.subagents.error": "失败", + "session.subagents.interrupted": "已中断", + "session.subagents.cancelled": "已取消", + "session.subagents.select": "选择子 Agent", + "session.subagents.open": "打开", + "session.subagents.panelError": "子 Agent 面板无法显示。", + "session.subagents.retry": "重试面板", "session.subagents.readonly": "该子 Agent 已完成，历史记录为只读。", "session.fork.derivedFrom": "从《{title}》派生", "session.fork.derivedFromUnknown": "从其他对话派生", @@ -947,10 +954,10 @@ export const dict = { "wiki.version": "v{{version}}", "wiki.links": "关联代码与文档", "wiki.stale": "已失效", - "wiki.type.knowledge": "知识", - "wiki.type.memory": "记忆", - "wiki.type.document": "文档", - "wiki.type.code": "代码", + "wiki.type.knowledge": "知识图", + "wiki.type.memory": "记忆图", + "wiki.type.document": "文档图", + "wiki.type.code": "代码图", "wiki.edit.button": "编辑", "wiki.edit.save": "保存", "wiki.edit.saved": "知识页已更新", @@ -1294,7 +1301,8 @@ export const dict = { "dialog.releaseNotes.media.alt": "发布预览", "toast.project.reloadFailed.title": "无法重新加载 {{project}}", "toast.project.rootRefused.title": "无法打开文件系统根目录", - "toast.project.rootRefused.description": "该对话指向文件系统根目录（“/”），出于安全考虑无法打开。这多半是遗留数据，请删除或改指到具体目录。", + "toast.project.rootRefused.description": + "该对话指向文件系统根目录（“/”），出于安全考虑无法打开。这多半是遗留数据，请删除或改指到具体目录。", "toast.project.rootRecoveryFailed.title": "无法恢复该无项目对话", "error.server.invalidConfiguration": "配置无效", "common.moreCountSuffix": " (还有 {{count}} 个)", @@ -1407,7 +1415,8 @@ export const dict = { "settings.about.product.title": "产品", "settings.about.product.description": "应用名称和版本。", "settings.about.attribution.title": "上游来源", - "settings.about.attribution.description": "DeepAgent Code 基于 MIT 许可的上游项目派生，并以 AGPL-3.0-or-later 发布。详见 NOTICE。", + "settings.about.attribution.description": + "DeepAgent Code 基于 MIT 许可的上游项目派生，并以 AGPL-3.0-or-later 发布。详见 NOTICE。", "settings.about.license.title": "许可证", "settings.general.section.advanced": "高级", "settings.general.row.shell.title": "终端 Shell", diff --git a/packages/app/src/i18n/zht.ts b/packages/app/src/i18n/zht.ts index 608fadcb..fb825dae 100644 --- a/packages/app/src/i18n/zht.ts +++ b/packages/app/src/i18n/zht.ts @@ -3,6 +3,13 @@ import { dict as en } from "./en" type Keys = keyof typeof en export const dict = { + "session.subagents.error": "失敗", + "session.subagents.interrupted": "已中斷", + "session.subagents.cancelled": "已取消", + "session.subagents.select": "選擇子 Agent", + "session.subagents.open": "開啟", + "session.subagents.panelError": "子 Agent 面板無法顯示。", + "session.subagents.retry": "重試面板", "command.category.suggested": "建議", "command.category.view": "檢視", "command.category.project": "專案", @@ -630,7 +637,8 @@ export const dict = { "oversight.takeover.unsupported": "接管記錄將在後端端點（P3.10）可用後啟用。", "oversight.takeover.failed": "失敗：{{error}}", "oversight.rollback.title": "回滾", - "oversight.rollback.description": "回滾某個工作階段中 Agent 產生的變更（透過 SessionRevert）。僅影響本工作區內的工作階段。此操作會回滾真實變更——請謹慎使用。", + "oversight.rollback.description": + "回滾某個工作階段中 Agent 產生的變更（透過 SessionRevert）。僅影響本工作區內的工作階段。此操作會回滾真實變更——請謹慎使用。", "oversight.rollback.sessionPlaceholder": "工作階段 ID（ses_…）", "oversight.rollback.reasonPlaceholder": "回滾原因…（可選）", "oversight.rollback.action": "回滾", @@ -741,10 +749,10 @@ export const dict = { "wiki.version": "v{{version}}", "wiki.links": "關聯程式碼與文件", "wiki.stale": "已失效", - "wiki.type.knowledge": "知識", - "wiki.type.memory": "記憶", - "wiki.type.document": "文件", - "wiki.type.code": "程式碼", + "wiki.type.knowledge": "知識圖", + "wiki.type.memory": "記憶圖", + "wiki.type.document": "文件圖", + "wiki.type.code": "程式碼圖", "wiki.edit.button": "編輯", "wiki.edit.save": "儲存", "wiki.edit.saved": "知識頁已更新", diff --git a/packages/app/src/pages/session/panel-error-boundary.tsx b/packages/app/src/pages/session/panel-error-boundary.tsx new file mode 100644 index 00000000..c5e1d104 --- /dev/null +++ b/packages/app/src/pages/session/panel-error-boundary.tsx @@ -0,0 +1,16 @@ +import { ErrorBoundary, type Component, type JSX } from "solid-js" + +export const PanelErrorBoundary: Component<{ + content: () => JSX.Element + onError: (error: unknown) => void + fallback: (retry: () => void) => JSX.Element +}> = (props) => ( + { + props.onError(error) + return props.fallback(reset) + }} + > + {props.content()} + +) diff --git a/packages/app/src/pages/session/session-side-panel.tsx b/packages/app/src/pages/session/session-side-panel.tsx index 1262d115..e8ccbef6 100644 --- a/packages/app/src/pages/session/session-side-panel.tsx +++ b/packages/app/src/pages/session/session-side-panel.tsx @@ -1,4 +1,16 @@ -import { For, Match, Show, Switch, createEffect, createMemo, createSignal, type ComponentProps, type JSX } from "solid-js" +import { + For, + Match, + Show, + Switch, + createEffect, + createMemo, + createSignal, + onMount, + type Component, + type ComponentProps, + type JSX, +} from "solid-js" import { createMediaQuery } from "@solid-primitives/media" import { Tabs } from "@deepagent-code/ui/tabs" import { IconButton } from "@deepagent-code/ui/icon-button" @@ -8,6 +20,7 @@ import { ResizeHandle } from "@deepagent-code/ui/resize-handle" import type { SnapshotFileDiff, VcsFileDiff } from "@deepagent-code/sdk/v2" import { RIGHT_PANEL_RAIL_PX, type RightPanelWidthBucket, type DockPanelID } from "@/context/layout" import { useTerminalHosts } from "@/context/terminal" +import { usePlatform } from "@/context/platform" import FileTree from "@/components/file-tree" import { SidePanelMcp } from "@/pages/session/side-panel-mcp" @@ -31,6 +44,7 @@ import { SidePanelProfile } from "@/pages/session/side-panel-profile" import { SidePanelIM } from "@/pages/session/side-panel-im" import { SidePanelDockHeader, SidePanelTerminal, SidePanelDebugConsole } from "@/pages/session/side-panel-terminal" import { ProblemsPanel } from "@/pages/session/problems-panel" +import { PanelErrorBoundary } from "@/pages/session/panel-error-boundary" type RenderDiff = (SnapshotFileDiff & { file: string }) | VcsFileDiff @@ -73,8 +87,22 @@ type PanelDef = { // Order = rail order. Groups are contiguous so the rail can draw a divider between them. const PANELS: readonly PanelDef[] = [ // Code - { mode: "review", icon: "review", titleKey: "session.tab.review", group: "code", bucket: "wide", keybind: "review.toggle" }, - { mode: "files", icon: "file-tree", titleKey: "settings.general.row.showFileTree.title", group: "code", bucket: "wide", keybind: "fileTree.toggle" }, + { + mode: "review", + icon: "review", + titleKey: "session.tab.review", + group: "code", + bucket: "wide", + keybind: "review.toggle", + }, + { + mode: "files", + icon: "file-tree", + titleKey: "settings.general.row.showFileTree.title", + group: "code", + bucket: "wide", + keybind: "fileTree.toggle", + }, // Agents — subagents is now the unified "子Agent监督" entry (Phase 2: oversight merged in). { mode: "subagents", icon: "agent-tree", titleKey: "session.subagents.title", group: "agents", bucket: "narrow" }, { mode: "im", icon: "bubble-5", titleKey: "session.tab.im", group: "agents", bucket: "narrow" }, @@ -87,8 +115,21 @@ const PANELS: readonly PanelDef[] = [ { mode: "debug", icon: "debug", titleKey: "session.panel.debug", group: "dev", bucket: "narrow" }, { mode: "profile", icon: "profile", titleKey: "session.panel.profile", group: "dev", bucket: "narrow" }, // Dock — the movable panels; only surface here when docked to the side (see gating below). - { mode: "terminal", icon: "terminal-active", titleKey: "session.panel.terminal", group: "dock", bucket: "narrow", keybind: "terminal.toggle" }, - { mode: "debug-console", icon: "code-lines", titleKey: "session.panel.debugConsole", group: "dock", bucket: "narrow" }, + { + mode: "terminal", + icon: "terminal-active", + titleKey: "session.panel.terminal", + group: "dock", + bucket: "narrow", + keybind: "terminal.toggle", + }, + { + mode: "debug-console", + icon: "code-lines", + titleKey: "session.panel.debugConsole", + group: "dock", + bucket: "narrow", + }, { mode: "problems", icon: "warning", titleKey: "session.panel.problems", group: "dock", bucket: "narrow" }, ] @@ -98,6 +139,11 @@ function renderDiff(value: SnapshotFileDiff | VcsFileDiff): value is RenderDiff return typeof value.file === "string" } +const GuardedSubagents: Component<{ onClose: () => void; onReady: () => void }> = (props) => { + onMount(props.onReady) + return +} + export function SessionSidePanel(props: { canReview: () => boolean diffs: () => (SnapshotFileDiff | VcsFileDiff)[] @@ -117,6 +163,7 @@ export function SessionSidePanel(props: { const debug = useDebug() const language = useLanguage() const command = useCommand() + const platform = usePlatform() const sync = useSync() const terminalHosts = useTerminalHosts() const { params, sessionKey, tabs, view } = useSessionLayout() @@ -135,12 +182,8 @@ export function SessionSidePanel(props: { const runningSubagentCount = createMemo( () => subagentChildren().filter((s) => sync.data.session_working(s.id)).length, ) - const interruptedSubagentCount = createMemo( - () => subagentChildren().filter(isSubagentInterrupted).length, - ) - const subagentAttentionCount = createMemo( - () => runningSubagentCount() + interruptedSubagentCount(), - ) + const interruptedSubagentCount = createMemo(() => subagentChildren().filter(isSubagentInterrupted).length) + const subagentAttentionCount = createMemo(() => runningSubagentCount() + interruptedSubagentCount()) // T3.1: the active panel mode (desktop only), derived once from the layout store instead of one memo // per panel. `isActive(mode)` and `open()` fall out of it. @@ -252,14 +295,21 @@ export function SessionSidePanel(props: { const commitNewItem = async () => { const s = newItemState() - if (!s || !s.name.trim()) { cancelNewItem(); return } + if (!s || !s.name.trim()) { + cancelNewItem() + return + } const name = s.name.trim() cancelNewItem() if (s.type === "file") { const res = await file.createFile(name) if (!res.ok) { const { showToast } = await import("@/utils/toast") - showToast({ variant: "error", title: language.t("session.files.create.failed"), description: res.error ?? "unknown" }) + showToast({ + variant: "error", + title: language.t("session.files.create.failed"), + description: res.error ?? "unknown", + }) } else { void file.load(name) } @@ -267,7 +317,11 @@ export function SessionSidePanel(props: { const res = await file.mkdir(name) if (!res.ok) { const { showToast } = await import("@/utils/toast") - showToast({ variant: "error", title: language.t("session.files.createFolder.failed"), description: res.error ?? "unknown" }) + showToast({ + variant: "error", + title: language.t("session.files.createFolder.failed"), + description: res.error ?? "unknown", + }) } } } @@ -306,6 +360,30 @@ export function SessionSidePanel(props: { // T3.2: a content panel's close button now CLOSES the panel (was: return to the menu list). The rail // stays visible, so the user re-opens with one click. const closePanel = () => view().rightPanel.close() + const panelBuild = platform.version ?? "dev" + let lastSubagentPanelError: unknown + let recoveringSubagentPanel = false + + const recordSubagentPanelError = (error: unknown) => { + if (lastSubagentPanelError === error) return + lastSubagentPanelError = error + recoveringSubagentPanel = true + view().rightPanel.reportFailure("subagents", panelBuild) + const text = + error instanceof Error + ? `${error.name}: ${error.message}${error.stack ? `\n${error.stack.slice(0, 4000)}` : ""}` + : String(error).slice(0, 4000) + console.error("ui.panel.render_failed", { mode: "subagents", route: location.href, build_hash: panelBuild, error }) + void platform + .recordFatalRendererError?.({ + error: `[panel:subagents] ${text}`, + url: location.href, + version: platform.version, + platform: platform.platform, + os: platform.os, + }) + .catch(() => undefined) + } // T3.1/T3.3: the rail entries, derived from PANELS — with a live badge count per panel. // review-change-count / subagent-attention-count feed `item.badge`; the rail renders whatever @@ -320,7 +398,11 @@ export function SessionSidePanel(props: { PANELS // Phase 3: terminal is side-native — always show in the rail. // Other movable dock panels (debug-console, problems) only appear when docked to the side. - .filter((p) => (MOVABLE_DOCK_MODES.includes(p.mode as DockPanelID) ? view().panel.location(p.mode as DockPanelID) === "side" : true)) + .filter((p) => + MOVABLE_DOCK_MODES.includes(p.mode as DockPanelID) + ? view().panel.location(p.mode as DockPanelID) === "side" + : true, + ) .map((p) => ({ ...p, title: language.t(p.titleKey), @@ -396,192 +478,237 @@ export function SessionSidePanel(props: { }} />
- - -
{props.reviewPanel()}
- - - - - - - {props.reviewCount()}{" "} - {language.t( - props.reviewCount() === 1 ? "session.review.change.one" : "session.review.change.other", - )} - - - {language.t("session.files.all")} - - - - - - - {language.t("common.loading")} - {language.t("common.loading.ellipsis")} -
- } - > + + +
{props.reviewPanel()}
+ + + + + + + {props.reviewCount()}{" "} + {language.t( + props.reviewCount() === 1 ? "session.review.change.one" : "session.review.change.other", + )} + + + {language.t("session.files.all")} + + + + + + + {language.t("common.loading")} + {language.t("common.loading.ellipsis")} +
+ } + > + props.focusReviewDiff(node.path)} + /> + + + + + + {/* V3.6 Phase 1B F5: file ops toolbar */} +
+ + {language.t("session.files.heading")} + + startNewItem("file")} + aria-label={language.t("session.files.newFile")} + /> + startNewItem("dir")} + aria-label={language.t("session.files.newFolder")} + /> +
+ + {(s) => ( +
+ setNewItemState({ ...s(), name: e.currentTarget.value })} + onKeyDown={(e) => { + if (e.key === "Enter") void commitNewItem() + if (e.key === "Escape") cancelNewItem() + }} + ref={(el) => setTimeout(() => el?.focus(), 0)} + /> + + +
+ )} + + + {empty(language.t("session.files.empty"))} + props.focusReviewDiff(node.path)} + onFileClick={(node) => openTab(file.tab(node.path))} /> - - - - - - {/* V3.6 Phase 1B F5: file ops toolbar */} -
- {language.t("session.files.heading")} - startNewItem("file")} - aria-label={language.t("session.files.newFile")} - /> - startNewItem("dir")} - aria-label={language.t("session.files.newFolder")} - /> -
- - {(s) => ( -
- setNewItemState({ ...s(), name: e.currentTarget.value })} - onKeyDown={(e) => { - if (e.key === "Enter") void commitNewItem() - if (e.key === "Escape") cancelNewItem() - }} - ref={(el) => setTimeout(() => el?.focus(), 0)} - /> - - -
- )} - - - {empty(language.t("session.files.empty"))} - - openTab(file.tab(node.path))} - /> - - - - -
- } - > - {(_p) => ( - { - const paused = debug.state.pausedLocation - return paused && paused.file === previewPath() ? paused.line : undefined - })()} - onToggleBreakpoint={(line) => { - const p = previewPath() - if (p) void debug.toggleBreakpoint(p, line) - }} - onNavigate={openFileAt} - gotoLine={gotoLine()} + + + + +
+ } + > + {(_p) => ( + { + const paused = debug.state.pausedLocation + return paused && paused.file === previewPath() ? paused.line : undefined + })()} + onToggleBreakpoint={(line) => { + const p = previewPath() + if (p) void debug.toggleBreakpoint(p, line) + }} + onNavigate={openFileAt} + gotoLine={gotoLine()} + /> + )} + + + + ( +
+

{language.t("session.subagents.panelError")}

+
+ + +
+
+ )} + content={() => ( + { + lastSubagentPanelError = undefined + view().rightPanel.clearFailure("subagents", panelBuild) + if (recoveringSubagentPanel) { + recoveringSubagentPanel = false + console.info("ui.panel.recovered", { mode: "subagents", build_hash: panelBuild }) + } + }} + /> + )} + /> + + + + + + + + + + + + + + + + + + + + + + + + view().rightPanel.close()} /> + + + view().panel.toggle("debug-console")} /> + + +
+ view().panel.toggle("problems")} /> - )} - - - - - - - - - - - - - - - - - - - - - - - - - - - - view().rightPanel.close()} /> - - - view().panel.toggle("debug-console")} /> - - -
- view().panel.toggle("problems")} /> -
- isActive("problems")} onOpenFile={openFileAt} /> +
+ isActive("problems")} onOpenFile={openFileAt} /> +
-
- - + +
diff --git a/packages/app/src/pages/session/side-panel-subagents-structure.test.ts b/packages/app/src/pages/session/side-panel-subagents-structure.test.ts new file mode 100644 index 00000000..369959e2 --- /dev/null +++ b/packages/app/src/pages/session/side-panel-subagents-structure.test.ts @@ -0,0 +1,56 @@ +import { expect, test } from "bun:test" +import { + ScriptKind, + ScriptTarget, + createSourceFile, + forEachChild, + isJsxAttribute, + isJsxElement, + isJsxExpression, + isJsxSelfClosingElement, + isStringLiteral, + type JsxOpeningLikeElement, + type Node, +} from "typescript" + +const nativeInteractive = new Set(["a", "button", "input", "select", "textarea", "summary"]) +const interactiveRoles = new Set(["button", "checkbox", "link", "menuitem", "option", "radio", "switch", "tab"]) + +function interactiveName(opening: JsxOpeningLikeElement) { + const tag = opening.tagName.getText() + if (nativeInteractive.has(tag)) return tag + const role = opening.attributes.properties.find( + (attribute) => isJsxAttribute(attribute) && attribute.name.getText() === "role", + ) + if (!role || !isJsxAttribute(role) || !role.initializer) return undefined + const value = isStringLiteral(role.initializer) + ? role.initializer.text + : isJsxExpression(role.initializer) && role.initializer.expression && isStringLiteral(role.initializer.expression) + ? role.initializer.expression.text + : undefined + return value && interactiveRoles.has(value) ? `[role=${value}]` : undefined +} + +test("subagent panel JSX has no nested interactive controls", async () => { + const path = `${import.meta.dir}/side-panel-subagents.tsx` + const source = createSourceFile(path, await Bun.file(path).text(), ScriptTarget.Latest, true, ScriptKind.TSX) + const violations: string[] = [] + + const visit = (node: Node, parents: string[]) => { + if (isJsxElement(node)) { + const current = interactiveName(node.openingElement) + if (current && parents.length > 0) violations.push(`${parents.join(" > ")} > ${current}`) + node.children.forEach((child) => visit(child, current ? [...parents, current] : parents)) + return + } + if (isJsxSelfClosingElement(node)) { + const current = interactiveName(node) + if (current && parents.length > 0) violations.push(`${parents.join(" > ")} > ${current}`) + return + } + forEachChild(node, (child) => visit(child, parents)) + } + + visit(source, []) + expect(violations).toEqual([]) +}) diff --git a/packages/app/src/pages/session/side-panel-subagents.tsx b/packages/app/src/pages/session/side-panel-subagents.tsx index 91a616fd..5ec133ad 100644 --- a/packages/app/src/pages/session/side-panel-subagents.tsx +++ b/packages/app/src/pages/session/side-panel-subagents.tsx @@ -6,7 +6,7 @@ import { useNavigate, useParams } from "@solidjs/router" import { useSDK } from "@/context/sdk" import { fetchCapabilities } from "@/components/deepagent/panel-goal.api" import { OversightDashboard } from "@/components/deepagent/oversight-dashboard" -import { isSubagentInterrupted } from "./subagent-state" +import { isSubagentInterrupted, subagentMetadata } from "./subagent-state" // Phase 2 (§3): SidePanelSubagents is now the single "子Agent监督" entry for the right rail. // It holds a `selectedSessionID` to track which subagent is being inspected; that selection @@ -41,12 +41,18 @@ export const SidePanelSubagents: Component<{ onClose: () => void }> = (props) => const oversightEnabled = () => capabilities()?.v4MultiAgentRuntime ?? false // ── subagent list ─────────────────────────────────────────────────────────── + const [persistedChildren] = createResource( + () => params.id, + async (sessionID) => (await sdk.client.session.children({ sessionID })).data ?? [], + ) const children = createMemo(() => { const id = params.id if (!id) return [] - return sync.data.session - .filter((s) => s.parentID === id) - .sort((a, b) => (b.time?.updated ?? 0) - (a.time?.updated ?? 0)) + return [ + ...new Map( + [...(persistedChildren() ?? []), ...sync.data.session.filter((s) => s.parentID === id)].map((s) => [s.id, s]), + ).values(), + ].sort((a, b) => (b.time?.updated ?? 0) - (a.time?.updated ?? 0)) }) // Three states. A subagent does one turn then finishes: the task tool persists a terminal marker @@ -54,28 +60,31 @@ export const SidePanelSubagents: Component<{ onClose: () => void }> = (props) => // "finished" (read-only) instead of "idle" (looks available). `session_working` is the live signal // for the brief window it's actually running; the persisted marker wins once set. // Phase 2 adds "interrupted" for subagents that were manually stopped. - const isFinished = (child: { metadata?: Record }): boolean => { - const sub = (child.metadata?.["deepagent"] as { subagent?: { finished?: boolean } } | undefined)?.subagent - return sub?.finished === true - } const isInterrupted = isSubagentInterrupted - const statusOf = ( - child: { id: string; metadata?: Record }, - ): "running" | "finished" | "interrupted" | "idle" => { + const statusOf = (child: { + id: string + metadata?: Record + }): "running" | "completed" | "error" | "interrupted" | "cancelled" | "idle" => { if (sync.data.session_working(child.id)) return "running" if (isInterrupted(child)) return "interrupted" - if (isFinished(child)) return "finished" + const state = subagentMetadata(child)?.state + if (state === "completed" || state === "error" || state === "cancelled") return state + if (subagentMetadata(child)?.finished === true) return "completed" return "idle" } - const statusLabel = (state: "running" | "finished" | "interrupted" | "idle") => + const statusLabel = (state: "running" | "completed" | "error" | "interrupted" | "cancelled" | "idle") => language.t( state === "running" ? "session.subagents.running" - : state === "finished" + : state === "completed" ? "session.subagents.finished" - : state === "interrupted" - ? "session.subagents.interrupted" - : "session.subagents.idle", + : state === "error" + ? "session.subagents.error" + : state === "interrupted" + ? "session.subagents.interrupted" + : state === "cancelled" + ? "session.subagents.cancelled" + : "session.subagents.idle", ) // ── selected session (監督対象) ─────────────────────────────────────────────── @@ -126,10 +135,11 @@ export const SidePanelSubagents: Component<{ onClose: () => void }> = (props) => {(child) => { const status = () => statusOf(child) + const reason = () => subagentMetadata(child)?.reason const isSelected = () => selectedSessionID() === child.id return (
void }> = (props) => {/* §3.4.2: click the row body to select; [Open] navigates. */} -
+ + + + + + {/* §3.4.3: sibling interactive control — never nest a button inside the row button. */} + ) diff --git a/packages/app/src/pages/session/subagent-state.test.ts b/packages/app/src/pages/session/subagent-state.test.ts index e5ff23d2..2e78f093 100644 --- a/packages/app/src/pages/session/subagent-state.test.ts +++ b/packages/app/src/pages/session/subagent-state.test.ts @@ -7,4 +7,12 @@ describe("isSubagentInterrupted", () => { expect(isSubagentInterrupted({ deepagent: { subagent: { interrupted: true } } })).toBe(true) expect(isSubagentInterrupted({ deepagent: { subagent: { state: "finished" } } })).toBe(false) }) + + test("reads metadata from a real session-shaped object", () => { + expect( + isSubagentInterrupted({ + metadata: { deepagent: { subagent: { state: "interrupted", reason: "human" } } }, + }), + ).toBe(true) + }) }) diff --git a/packages/app/src/pages/session/subagent-state.ts b/packages/app/src/pages/session/subagent-state.ts index 188ceafd..06079607 100644 --- a/packages/app/src/pages/session/subagent-state.ts +++ b/packages/app/src/pages/session/subagent-state.ts @@ -1,10 +1,29 @@ type SubagentMetadata = { + finished?: boolean state?: string + reason?: string interrupted?: boolean } -/** Supports durable state markers and legacy boolean interruption markers. */ -export const isSubagentInterrupted = (metadata?: Record) => { - const subagent = (metadata?.["deepagent"] as { subagent?: SubagentMetadata } | undefined)?.subagent +const isRecord = (value: unknown): value is Record => + typeof value === "object" && value !== null && !Array.isArray(value) + +export const subagentMetadata = (value?: unknown) => { + if (!isRecord(value)) return undefined + const metadata = isRecord(value.metadata) ? value.metadata : value + const deepagent = isRecord(metadata.deepagent) ? metadata.deepagent : undefined + const subagent = isRecord(deepagent?.subagent) ? deepagent.subagent : undefined + if (!subagent) return undefined + return { + ...(typeof subagent.finished === "boolean" ? { finished: subagent.finished } : {}), + ...(typeof subagent.state === "string" ? { state: subagent.state } : {}), + ...(typeof subagent.reason === "string" ? { reason: subagent.reason } : {}), + ...(typeof subagent.interrupted === "boolean" ? { interrupted: subagent.interrupted } : {}), + } satisfies SubagentMetadata +} + +/** Supports durable state markers, real Session records, and legacy boolean interruption markers. */ +export const isSubagentInterrupted = (value?: unknown) => { + const subagent = subagentMetadata(value) return subagent?.state === "interrupted" || subagent?.interrupted === true } diff --git a/packages/core/src/database/migration.gen.ts b/packages/core/src/database/migration.gen.ts index b8ff85ff..74ba0956 100644 --- a/packages/core/src/database/migration.gen.ts +++ b/packages/core/src/database/migration.gen.ts @@ -54,5 +54,6 @@ export const migrations = ( import("./migration/20260712050000_session_steer_queue"), import("./migration/20260719000000_deepagent_consumer_group"), import("./migration/20260722000000_session_steer_correlation"), + import("./migration/20260724134000_task_run_delivery"), ]) ).map((module) => module.default) satisfies DatabaseMigration.Migration[] diff --git a/packages/core/src/database/migration/20260724134000_task_run_delivery.ts b/packages/core/src/database/migration/20260724134000_task_run_delivery.ts new file mode 100644 index 00000000..95cd5318 --- /dev/null +++ b/packages/core/src/database/migration/20260724134000_task_run_delivery.ts @@ -0,0 +1,92 @@ +import { Effect } from "effect" +import type { DatabaseMigration } from "../migration" + +export default { + id: "20260724134000_task_run_delivery", + up(tx) { + return Effect.gen(function* () { + yield* tx.run(` + CREATE TABLE IF NOT EXISTS task_run ( + run_id TEXT PRIMARY KEY, + root_run_id TEXT, + request_hash TEXT NOT NULL, + parent_session_id TEXT NOT NULL REFERENCES session(id) ON DELETE CASCADE, + parent_message_id TEXT NOT NULL, + tool_call_id TEXT NOT NULL, + child_session_id TEXT NOT NULL, + generation INTEGER NOT NULL, + delivery_mode TEXT NOT NULL CHECK (delivery_mode IN ('foreground', 'background')), + phase TEXT NOT NULL CHECK (phase IN ('admission', 'research', 'finalize', 'settled')), + state TEXT NOT NULL CHECK (state IN ('admitted', 'provisioning', 'researching', 'finalizing', 'completed', 'error', 'cancelled', 'interrupted')), + reason TEXT, + attempts INTEGER NOT NULL DEFAULT 0, + execution_owner TEXT, + lease_expires_at INTEGER, + raw_result_message_id TEXT, + structured_result_message_id TEXT, + output TEXT, + error TEXT, + time_created INTEGER NOT NULL, + time_updated INTEGER NOT NULL, + time_settled INTEGER + ) + `) + yield* tx.run(` + CREATE UNIQUE INDEX IF NOT EXISTS task_run_child_generation_idx + ON task_run (child_session_id, generation) + `) + yield* tx.run(` + CREATE UNIQUE INDEX IF NOT EXISTS task_run_child_active_idx + ON task_run (child_session_id) + WHERE state IN ('admitted', 'provisioning', 'researching', 'finalizing') + `) + yield* tx.run(` + CREATE INDEX IF NOT EXISTS task_run_parent_state_idx + ON task_run (parent_session_id, state, time_updated) + `) + yield* tx.run(` + CREATE INDEX IF NOT EXISTS task_run_root_idx + ON task_run (root_run_id) + `) + yield* tx.run(` + CREATE TABLE IF NOT EXISTS task_admission ( + admission_key TEXT PRIMARY KEY, + request_hash TEXT NOT NULL, + run_id TEXT NOT NULL REFERENCES task_run(run_id) ON DELETE CASCADE, + parent_session_id TEXT NOT NULL REFERENCES session(id) ON DELETE CASCADE, + parent_message_id TEXT NOT NULL, + tool_call_id TEXT NOT NULL, + delivery_mode TEXT NOT NULL CHECK (delivery_mode IN ('foreground', 'background')), + time_created INTEGER NOT NULL + ) + `) + yield* tx.run(` + CREATE INDEX IF NOT EXISTS task_admission_run_idx + ON task_admission (run_id) + `) + yield* tx.run(` + CREATE TABLE IF NOT EXISTS task_notification_outbox ( + id TEXT PRIMARY KEY, + run_id TEXT NOT NULL UNIQUE REFERENCES task_run(run_id) ON DELETE CASCADE, + message_id TEXT NOT NULL UNIQUE, + parent_session_id TEXT NOT NULL REFERENCES session(id) ON DELETE CASCADE, + directory TEXT NOT NULL, + payload TEXT NOT NULL, + status TEXT NOT NULL CHECK (status IN ('pending', 'delivering', 'delivered', 'dead')), + attempts INTEGER NOT NULL DEFAULT 0, + available_at INTEGER NOT NULL, + lease_owner TEXT, + lease_expires_at INTEGER, + last_error TEXT, + time_created INTEGER NOT NULL, + time_updated INTEGER NOT NULL, + time_delivered INTEGER + ) + `) + yield* tx.run(` + CREATE INDEX IF NOT EXISTS task_notification_outbox_due_idx + ON task_notification_outbox (status, available_at, lease_expires_at) + `) + }) + }, +} satisfies DatabaseMigration.Migration diff --git a/packages/core/src/session/sql.ts b/packages/core/src/session/sql.ts index c509d4a0..1bdd4971 100644 --- a/packages/core/src/session/sql.ts +++ b/packages/core/src/session/sql.ts @@ -13,6 +13,7 @@ import { WorkspaceV2 } from "../workspace" import { Timestamps } from "../database/schema.sql" import type { SystemContext } from "../system-context/index" import { AgentV2 } from "../agent" +import { sql } from "drizzle-orm" type SessionMessageData = Omit<(typeof SessionMessage.Message)["Encoded"], "type" | "id"> type V1MessageData = Omit @@ -219,3 +220,100 @@ export const SessionContextEpochTable = sqliteTable("session_context_epoch", { replacement_seq: integer(), revision: integer().notNull().default(0), }) + +export const TaskRunTable = sqliteTable( + "task_run", + { + run_id: text().primaryKey(), + root_run_id: text(), + request_hash: text().notNull(), + parent_session_id: text() + .$type() + .notNull() + .references(() => SessionTable.id, { onDelete: "cascade" }), + parent_message_id: text().$type().notNull(), + tool_call_id: text().notNull(), + child_session_id: text().$type().notNull(), + generation: integer().notNull(), + delivery_mode: text().$type<"foreground" | "background">().notNull(), + phase: text().$type<"admission" | "research" | "finalize" | "settled">().notNull(), + state: text() + .$type< + "admitted" | "provisioning" | "researching" | "finalizing" | "completed" | "error" | "cancelled" | "interrupted" + >() + .notNull(), + reason: text(), + attempts: integer().notNull().default(0), + execution_owner: text(), + lease_expires_at: integer(), + raw_result_message_id: text().$type(), + structured_result_message_id: text().$type(), + output: text(), + error: text({ mode: "json" }).$type<{ code: string; message: string; data?: Record }>(), + time_created: integer().notNull(), + time_updated: integer().notNull(), + time_settled: integer(), + }, + (table) => [ + uniqueIndex("task_run_child_generation_idx").on(table.child_session_id, table.generation), + uniqueIndex("task_run_child_active_idx") + .on(table.child_session_id) + .where(sql`${table.state} IN ('admitted', 'provisioning', 'researching', 'finalizing')`), + index("task_run_parent_state_idx").on(table.parent_session_id, table.state, table.time_updated), + index("task_run_root_idx").on(table.root_run_id), + ], +) + +export const TaskAdmissionTable = sqliteTable( + "task_admission", + { + admission_key: text().primaryKey(), + request_hash: text().notNull(), + run_id: text() + .notNull() + .references(() => TaskRunTable.run_id, { onDelete: "cascade" }), + parent_session_id: text() + .$type() + .notNull() + .references(() => SessionTable.id, { onDelete: "cascade" }), + parent_message_id: text().$type().notNull(), + tool_call_id: text().notNull(), + delivery_mode: text().$type<"foreground" | "background">().notNull(), + time_created: integer().notNull(), + }, + (table) => [index("task_admission_run_idx").on(table.run_id)], +) + +export const TaskNotificationOutboxTable = sqliteTable( + "task_notification_outbox", + { + id: text().primaryKey(), + run_id: text() + .notNull() + .unique() + .references(() => TaskRunTable.run_id, { onDelete: "cascade" }), + message_id: text().$type().notNull().unique(), + parent_session_id: text() + .$type() + .notNull() + .references(() => SessionTable.id, { onDelete: "cascade" }), + directory: DatabasePath.directoryColumn().notNull(), + payload: text({ mode: "json" }) + .$type<{ + agent: string + variant?: string + text: string + }>() + .notNull(), + status: text().$type<"pending" | "delivering" | "delivered" | "dead">().notNull(), + attempts: integer().notNull().default(0), + available_at: integer().notNull(), + lease_owner: text(), + lease_expires_at: integer(), + last_error: text(), + time_created: integer().notNull(), + time_updated: integer().notNull(), + time_delivered: integer(), + }, + (table) => [index("task_notification_outbox_due_idx").on(table.status, table.available_at, table.lease_expires_at)], +) diff --git a/packages/core/src/v1/session.ts b/packages/core/src/v1/session.ts index 48d4556a..7fdc9ee9 100644 --- a/packages/core/src/v1/session.ts +++ b/packages/core/src/v1/session.ts @@ -39,6 +39,18 @@ export const StructuredOutputError = NamedError.create("StructuredOutputError", message: Schema.String, retries: NonNegativeInt, }) +export const DoomLoopError = NamedError.create("DoomLoopError", { + message: Schema.String, + tool: Schema.String, + period: NonNegativeInt, + count: NonNegativeInt, +}) +export const TaskBudgetExceededError = NamedError.create("TaskBudgetExceededError", { + message: Schema.String, + budget: Schema.Literals(["steps", "tokens", "wall_time", "no_progress"]), + limit: NonNegativeInt, + used: NonNegativeInt, +}) export const APIError = NamedError.create("APIError", { message: Schema.String, statusCode: Schema.optional(NonNegativeInt), @@ -394,6 +406,8 @@ const AssistantErrorSchema = Schema.Union([ OutputLengthError.EffectSchema, AbortedError.EffectSchema, StructuredOutputError.EffectSchema, + DoomLoopError.EffectSchema, + TaskBudgetExceededError.EffectSchema, ContextOverflowError.EffectSchema, APIError.EffectSchema, OutputDegenerationError.EffectSchema, diff --git a/packages/core/test/database-migration.test.ts b/packages/core/test/database-migration.test.ts index d7d0b3b0..b0d82bdb 100644 --- a/packages/core/test/database-migration.test.ts +++ b/packages/core/test/database-migration.test.ts @@ -74,6 +74,20 @@ describe("DatabaseMigration", () => { sql`SELECT name, dflt_value FROM pragma_table_info('session_context_epoch') WHERE name = 'agent'`, ), ).toEqual({ name: "agent", dflt_value: "'build'" }) + expect( + yield* db.all( + sql`SELECT name FROM sqlite_master WHERE type = 'table' AND name IN ('task_run', 'task_admission', 'task_notification_outbox') ORDER BY name`, + ), + ).toEqual([{ name: "task_admission" }, { name: "task_notification_outbox" }, { name: "task_run" }]) + expect( + yield* db.all( + sql`SELECT name FROM sqlite_master WHERE type = 'index' AND name IN ('task_run_child_generation_idx', 'task_run_child_active_idx', 'task_notification_outbox_due_idx') ORDER BY name`, + ), + ).toEqual([ + { name: "task_notification_outbox_due_idx" }, + { name: "task_run_child_active_idx" }, + { name: "task_run_child_generation_idx" }, + ]) expect(yield* db.get(sql`SELECT count(*) as count FROM migration`)).toEqual({ count: migrations.length }) expect( yield* db.all( diff --git a/packages/deepagent-code/package.json b/packages/deepagent-code/package.json index a2ebf635..2cac42ec 100644 --- a/packages/deepagent-code/package.json +++ b/packages/deepagent-code/package.json @@ -115,6 +115,7 @@ "@zip.js/zip.js": "2.7.62", "ai": "catalog:", "ai-gateway-provider": "3.1.2", + "ajv": "8.20.0", "bonjour-service": "1.3.0", "chokidar": "4.0.3", "cross-spawn": "catalog:", diff --git a/packages/deepagent-code/src/agent/agent.ts b/packages/deepagent-code/src/agent/agent.ts index 66d856d0..78079263 100644 --- a/packages/deepagent-code/src/agent/agent.ts +++ b/packages/deepagent-code/src/agent/agent.ts @@ -173,7 +173,8 @@ export const layer = Layer.effect( // autonomously sets the objective, produces design/plan as needed, and executes it end-to-end. auto: { name: "auto", - description: "Autonomous mode. The agent sets the objective, designs and plans as needed, then executes to completion.", + description: + "Autonomous mode. The agent sets the objective, designs and plans as needed, then executes to completion.", options: {}, permission: Permission.merge( defaults, @@ -313,6 +314,7 @@ export const layer = Layer.effect( external_directory: readonlyExternalDirectory, }), user, + Permission.fromConfig({ doom_loop: "ask" }), ), description: `Deep sub-module research agent. Use this when you need a decidable explanation of HOW a specific sub-module or subsystem works (not just where it is): its mechanism, key files, outward interfaces, risks, and open questions. Prefer this over "explore" when the task is to understand and report on one module in depth so you can synthesize a plan; prefer "explore" for quick file/keyword location. Returns a structured research result.`, prompt: PROMPT_RESEARCHER, @@ -335,6 +337,7 @@ export const layer = Layer.effect( external_directory: readonlyExternalDirectory, }), user, + Permission.fromConfig({ doom_loop: "ask" }), ), description: `Independent, adversarial review agent. Use this to critique a plan or a set of changes from a skeptical, outside perspective — its default stance is that the change has problems. It hunts for correctness bugs, security issues, edge cases, convention conflicts, and missing tests, and reports reproducible failure scenarios. Read-only. Returns structured findings with an overall verdict.`, prompt: PROMPT_REVIEWER, @@ -498,14 +501,7 @@ export const layer = Layer.effect( const list = Effect.fnUntraced(function* () { const cfg = yield* config.get() const preferred = cfg.default_agent ? canonicalAgentName(cfg.default_agent) : "auto" - return pipe( - agents, - values(), - sortBy( - [(x) => x.name === preferred, "desc"], - [(x) => x.name, "asc"], - ), - ) + return pipe(agents, values(), sortBy([(x) => x.name === preferred, "desc"], [(x) => x.name, "asc"])) }) const defaultInfo = Effect.fnUntraced(function* () { diff --git a/packages/deepagent-code/src/effect/instance-registry.ts b/packages/deepagent-code/src/effect/instance-registry.ts index 59c556e0..a29ef91e 100644 --- a/packages/deepagent-code/src/effect/instance-registry.ts +++ b/packages/deepagent-code/src/effect/instance-registry.ts @@ -1,4 +1,24 @@ +import type { InstanceContext } from "@/project/instance-context" + const disposers = new Set<(directory: string) => Promise>() +const initializers = new Set<(context: InstanceContext) => Promise>() + +export function registerInitializer(initializer: (context: InstanceContext) => Promise) { + initializers.add(initializer) + return () => { + initializers.delete(initializer) + } +} + +export async function initializeInstance(context: InstanceContext) { + const results = await Promise.allSettled([...initializers].map((initializer) => initializer(context))) + const failed = results.filter((result): result is PromiseRejectedResult => result.status === "rejected") + if (failed.length > 0) + throw new AggregateError( + failed.map((result) => result.reason), + "Instance initialization failed", + ) +} export function registerDisposer(disposer: (directory: string) => Promise) { disposers.add(disposer) diff --git a/packages/deepagent-code/src/effect/runtime-flags.ts b/packages/deepagent-code/src/effect/runtime-flags.ts index 45c100d6..311721fb 100644 --- a/packages/deepagent-code/src/effect/runtime-flags.ts +++ b/packages/deepagent-code/src/effect/runtime-flags.ts @@ -57,6 +57,10 @@ export class Service extends ConfigService.Service()("@deepagent-code/R // v4.0.4 块1 (I33-4): 子 Agent 结果注入父会话的有界长度(字符数)。超过则父只收截断摘要 + 指向子 // session 的引用(全量 text 不丢,仍在子 session 可查)。默认 undefined = 全量注入(逐字节等价现状)。 subagentOutputMaxChars: positiveInteger("DEEPAGENT_CODE_SUBAGENT_OUTPUT_MAX_CHARS"), + subagentResearchStepLimit: positiveInteger("DEEPAGENT_CODE_SUBAGENT_RESEARCH_STEP_LIMIT"), + subagentResearchTokenLimit: positiveInteger("DEEPAGENT_CODE_SUBAGENT_RESEARCH_TOKEN_LIMIT"), + subagentResearchWallMs: positiveInteger("DEEPAGENT_CODE_SUBAGENT_RESEARCH_WALL_MS"), + subagentNoProgressLimit: positiveInteger("DEEPAGENT_CODE_SUBAGENT_NO_PROGRESS_LIMIT"), experimentalLspTy: bool("DEEPAGENT_CODE_EXPERIMENTAL_LSP_TY"), experimentalLspTool: enabledByExperimental("DEEPAGENT_CODE_EXPERIMENTAL_LSP_TOOL"), // V3.8 App-A C2.5 (Stage 5): query_log tool — lets the agent retrieve slices of the append-only diff --git a/packages/deepagent-code/src/project/instance-store.ts b/packages/deepagent-code/src/project/instance-store.ts index 26db856c..e8ac844a 100644 --- a/packages/deepagent-code/src/project/instance-store.ts +++ b/packages/deepagent-code/src/project/instance-store.ts @@ -2,7 +2,7 @@ import { GlobalBus } from "@/bus/global" import { serviceUse } from "@deepagent-code/core/effect/service-use" import { WorkspaceContext } from "@/control-plane/workspace-context" import { InstanceRef } from "@/effect/instance-ref" -import { disposeInstance as runDisposers } from "@/effect/instance-registry" +import { disposeInstance as runDisposers, initializeInstance } from "@/effect/instance-registry" import { FSUtil } from "@deepagent-code/core/fs-util" import { Context, Deferred, Duration, Effect, Exit, Layer, Scope } from "effect" import { assertSafeInstanceRoot, type InstanceContext } from "./instance-context" @@ -63,6 +63,13 @@ export const layer: Layer.Layer initializeInstance(ctx)).pipe( + Effect.catchCause((cause) => + Effect.logWarning("instance initializer failed").pipe( + Effect.annotateLogs({ directory: ctx.directory, cause }), + ), + ), + ) return ctx }).pipe(Effect.withSpan("InstanceStore.boot")) diff --git a/packages/deepagent-code/src/session/message-v2.ts b/packages/deepagent-code/src/session/message-v2.ts index cacb518f..7b5210a5 100644 --- a/packages/deepagent-code/src/session/message-v2.ts +++ b/packages/deepagent-code/src/session/message-v2.ts @@ -9,12 +9,14 @@ import { AuthError, CompactionPart, ContextOverflowError, + DoomLoopError, Info, OutputDegenerationError, OutputLengthError, Part, StructuredOutputError, SubtaskPart, + TaskBudgetExceededError, User, WithParts, type ToolPart, @@ -162,6 +164,13 @@ function providerMeta(metadata: Record | undefined) { return Object.keys(rest).length > 0 ? rest : undefined } +function toolCallProviderMeta(metadata: Record | undefined, differentModel: boolean) { + const type = metadata?.deepagent?.toolType + if (type !== "custom" && type !== "function") return differentModel ? undefined : providerMeta(metadata) + if (!differentModel) return providerMeta(metadata) + return { deepagent: { toolType: type } } +} + export const toModelMessagesEffect = Effect.fnUntraced(function* ( input: WithParts[], model: Provider.Model, @@ -353,7 +362,9 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* ( input: part.state.input, output, ...(part.metadata?.providerExecuted ? { providerExecuted: true } : {}), - ...(differentModel ? {} : { callProviderMetadata: providerMeta(part.metadata) }), + ...(toolCallProviderMeta(part.metadata, differentModel) + ? { callProviderMetadata: toolCallProviderMeta(part.metadata, differentModel) } + : {}), }) } if (part.state.status === "error") { @@ -366,7 +377,9 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* ( input: part.state.input, output, ...(part.metadata?.providerExecuted ? { providerExecuted: true } : {}), - ...(differentModel ? {} : { callProviderMetadata: providerMeta(part.metadata) }), + ...(toolCallProviderMeta(part.metadata, differentModel) + ? { callProviderMetadata: toolCallProviderMeta(part.metadata, differentModel) } + : {}), }) } else { assistantMessage.parts.push({ @@ -376,7 +389,9 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* ( input: part.state.input, errorText: part.state.error, ...(part.metadata?.providerExecuted ? { providerExecuted: true } : {}), - ...(differentModel ? {} : { callProviderMetadata: providerMeta(part.metadata) }), + ...(toolCallProviderMeta(part.metadata, differentModel) + ? { callProviderMetadata: toolCallProviderMeta(part.metadata, differentModel) } + : {}), }) } } @@ -390,7 +405,9 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* ( input: part.state.input, errorText: "[Tool execution was interrupted]", ...(part.metadata?.providerExecuted ? { providerExecuted: true } : {}), - ...(differentModel ? {} : { callProviderMetadata: providerMeta(part.metadata) }), + ...(toolCallProviderMeta(part.metadata, differentModel) + ? { callProviderMetadata: toolCallProviderMeta(part.metadata, differentModel) } + : {}), }) } if (part.type === "reasoning") { @@ -656,6 +673,10 @@ export function fromError( return e case OutputDegenerationError.isInstance(e): return e + case DoomLoopError.isInstance(e): + return e + case TaskBudgetExceededError.isInstance(e): + return e case LoadAPIKeyError.isInstance(e): return new AuthError( { diff --git a/packages/deepagent-code/src/session/processor.ts b/packages/deepagent-code/src/session/processor.ts index 3aca0ed8..346bcacd 100644 --- a/packages/deepagent-code/src/session/processor.ts +++ b/packages/deepagent-code/src/session/processor.ts @@ -33,6 +33,7 @@ import * as DateTime from "effect/DateTime" import { RuntimeFlags } from "@/effect/runtime-flags" import { toolFileSourceFromUri, Usage, type LLMEvent } from "@deepagent-code/llm" import { ToolOutput } from "@deepagent-code/core/tool-output" +import { AgentGateway } from "@deepagent-code/core/agent-gateway" const DOOM_LOOP_THRESHOLD = 3 const DOOM_LOOP_SEQUENCE_WINDOW = 12 @@ -80,6 +81,23 @@ export class ToolSequenceTracker { private readonly calls: { fingerprint: string; done: boolean }[] = [] private readonly callIdToIndex = new Map() private readonly triggeredSequences = new Set() + private fingerprintResolver: ((toolName: string, input: unknown) => unknown) | undefined + private resultFingerprintResolver: ((toolName: string, result: unknown) => unknown) | undefined + private previousResultSignature: string | undefined + private previousProgressSignature: string | undefined + private equivalentResultCount = 0 + + setFingerprintResolver(resolver: (toolName: string, input: unknown) => unknown): void { + this.fingerprintResolver = resolver + } + + setResultFingerprintResolver(resolver: (toolName: string, result: unknown) => unknown): void { + this.resultFingerprintResolver = resolver + } + + fingerprint(toolName: string, input: unknown): string { + return toolFingerprint(toolName, this.fingerprintResolver ? this.fingerprintResolver(toolName, input) : input) + } /** Record a newly started (running) tool call. */ push(callId: string, fingerprint: string): void { @@ -101,12 +119,33 @@ export class ToolSequenceTracker { * settleToolCall so that the "prior calls must be done" invariant holds * before the next tool starts. */ - markDone(callId: string): void { + markDone( + callId: string, + toolName?: string, + result?: unknown, + progress?: { snapshot: string | undefined; plan: unknown }, + ): { count: number } | undefined { const idx = this.callIdToIndex.get(callId) if (idx !== undefined && idx >= 0 && idx < this.calls.length) { this.calls[idx].done = true } this.callIdToIndex.delete(callId) + const resolved = toolName && this.resultFingerprintResolver?.(toolName, result) + if (idx === undefined || resolved === undefined) { + this.previousResultSignature = undefined + this.previousProgressSignature = undefined + this.equivalentResultCount = 0 + return undefined + } + const resultSignature = `${toolName}:${canonicalJson(resolved)}` + const progressSignature = canonicalJson(progress) + this.equivalentResultCount = + resultSignature === this.previousResultSignature && progressSignature === this.previousProgressSignature + ? this.equivalentResultCount + 1 + : 1 + this.previousResultSignature = resultSignature + this.previousProgressSignature = progressSignature + return { count: this.equivalentResultCount } } /** @@ -173,13 +212,13 @@ export class ToolSequenceTracker { // Detects repetitive/stuck output before it grows unbounded; configurable via // RuntimeFlags.degenerationDetectorMode ("off" | "shadow" | "enforce"). const DEGENERATION_DETECTOR_VERSION = "1.0" -const DEGENERATION_ENABLE_THRESHOLD = 20_000 // chars before detection starts -const DEGENERATION_WINDOW_SIZE = 4_000 // sliding window width in chars -const DEGENERATION_SAMPLE_INTERVAL = 500 // chars between samples -const DEGENERATION_N = 4 // N-gram size -const DEGENERATION_RATIO_THRESHOLD = 0.70 // repeated N-gram fraction +const DEGENERATION_ENABLE_THRESHOLD = 20_000 // chars before detection starts +const DEGENERATION_WINDOW_SIZE = 4_000 // sliding window width in chars +const DEGENERATION_SAMPLE_INTERVAL = 500 // chars between samples +const DEGENERATION_N = 4 // N-gram size +const DEGENERATION_RATIO_THRESHOLD = 0.7 // repeated N-gram fraction const DEGENERATION_SIMILARITY_THRESHOLD = 0.85 // Jaccard threshold between windows -const DEGENERATION_K = 3 // consecutive samples required +const DEGENERATION_K = 3 // consecutive samples required class DegenerationDetector { private totalChars = 0 @@ -227,9 +266,7 @@ class DegenerationDetector { // Maintain sliding window: keep only the last WINDOW_SIZE chars const combined = this.windowText + delta this.windowText = - combined.length > DEGENERATION_WINDOW_SIZE - ? combined.slice(combined.length - DEGENERATION_WINDOW_SIZE) - : combined + combined.length > DEGENERATION_WINDOW_SIZE ? combined.slice(combined.length - DEGENERATION_WINDOW_SIZE) : combined if (this.totalChars < DEGENERATION_ENABLE_THRESHOLD) return { triggered: false } if (this.charsSinceLastSample < DEGENERATION_SAMPLE_INTERVAL) return { triggered: false } @@ -268,7 +305,7 @@ export interface Handle { output: string attachments?: SessionV1.FilePart[] }, - ) => Effect.Effect + ) => Effect.Effect readonly process: (streamInput: LLM.StreamInput) => Effect.Effect } @@ -283,6 +320,8 @@ type Input = { * Absent only in legacy callers that have not been updated yet. */ sequenceTracker?: ToolSequenceTracker + loopPolicy?: "ask" | "error" + noProgressLimit?: number } export interface Interface { @@ -365,13 +404,19 @@ export const layer = Layer.effect( aborted, }) - const settleToolCall = Effect.fn("SessionProcessor.settleToolCall")(function* (toolCallID: string) { + const settleToolCall = Effect.fn("SessionProcessor.settleToolCall")(function* ( + toolCallID: string, + toolName?: string, + result?: unknown, + progress?: { snapshot: string | undefined; plan: unknown }, + ) { // Notify the activity-level tracker that this call has finished so it // satisfies the "prior calls must be done" precondition for detection. - ctx.sequenceTracker?.markDone(toolCallID) + const noProgress = ctx.sequenceTracker?.markDone(toolCallID, toolName, result, progress) const done = ctx.toolcalls[toolCallID]?.done delete ctx.toolcalls[toolCallID] if (done) yield* Deferred.succeed(done, undefined).pipe(Effect.ignore) + return noProgress }) const ensureV2AssistantMessage = Effect.fn("SessionProcessor.ensureV2AssistantMessage")(function* () { @@ -456,7 +501,33 @@ export const layer = Layer.effect( attachments: output.attachments, }, }) - yield* settleToolCall(toolCallID) + const noProgress = yield* settleToolCall( + toolCallID, + match.part.tool, + output, + input.noProgressLimit + ? { + snapshot: yield* snapshot.track(), + plan: AgentGateway.DeepAgentSessionState.getPlan(ctx.sessionID), + } + : undefined, + ) + if (input.noProgressLimit && noProgress && noProgress.count >= input.noProgressLimit) { + slog.warn("subagent.loop.detected", { + fingerprint_kind: "tool_result", + period: 1, + count: noProgress.count, + tool: match.part.tool, + }) + yield* Effect.fail( + new SessionV1.TaskBudgetExceededError({ + message: `Non-interactive activity stopped after ${noProgress.count} equivalent ${match.part.tool} results without observable progress.`, + budget: "no_progress", + limit: input.noProgressLimit, + used: noProgress.count, + }), + ) + } }) const failToolCall = Effect.fn("SessionProcessor.failToolCall")(function* (toolCallID: string, error: unknown) { @@ -628,9 +699,7 @@ export const layer = Layer.effect( // Summary (compaction) processors are excluded — they are short-lived // and use a distinct reasoning style that should never be circuit-broken. if (!ctx.assistantMessage.summary) { - ctx.degenerationDetectors[value.id] = new DegenerationDetector( - flags.degenerationDetectorMode, - ) + ctx.degenerationDetectors[value.id] = new DegenerationDetector(flags.degenerationDetectorMode) } yield* session.updatePart(ctx.reasoningMap[value.id]) return @@ -785,10 +854,26 @@ export const layer = Layer.effect( // F1: Activity-level cross-message loop detection (primary path) // --------------------------------------------------------------- if (ctx.sequenceTracker) { - const fp = toolFingerprint(value.name, input) + const fp = ctx.sequenceTracker.fingerprint(value.name, input) ctx.sequenceTracker.push(value.id, fp) const detected = ctx.sequenceTracker.detect() if (detected && !ctx.sequenceTracker.hasTriggered(detected.sequenceKey)) { + slog.warn("subagent.loop.detected", { + fingerprint_kind: "tool_input_sequence", + period: detected.period, + count: detected.count, + tool: value.name, + }) + if (ctx.loopPolicy === "error") { + return yield* Effect.fail( + new SessionV1.DoomLoopError({ + message: `Non-interactive activity stopped after a repeated ${value.name} tool sequence was detected.`, + tool: value.name, + period: detected.period, + count: detected.count, + }), + ) + } const agent = yield* agents.get(ctx.assistantMessage.agent) yield* permission.ask({ permission: "doom_loop", @@ -831,15 +916,23 @@ export const layer = Layer.effect( !singleRepeat && detectRepeatingSequence( parts - .filter( - (part): part is SessionV1.ToolPart => - part.type === "tool" && part.state.status !== "pending", - ) + .filter((part): part is SessionV1.ToolPart => part.type === "tool" && part.state.status !== "pending") .map((part) => `${part.tool}:${JSON.stringify(part.state.input)}`), ) if (!singleRepeat && !sequenceRepeat) return + if (ctx.loopPolicy === "error") { + return yield* Effect.fail( + new SessionV1.DoomLoopError({ + message: `Non-interactive activity stopped after a repeated ${value.name} tool sequence was detected.`, + tool: value.name, + period: sequenceRepeat ? 2 : 1, + count: DOOM_LOOP_MIN_REPEATS, + }), + ) + } + const agent = yield* agents.get(ctx.assistantMessage.agent) yield* permission.ask({ permission: "doom_loop", @@ -1006,9 +1099,9 @@ export const layer = Layer.effect( // Response-side prompt-cache monitor: compare this step's real cache-read ratio to the // previous step and warn if it collapsed while the prompt didn't shrink (suspected cache // break the static system-hash tripwire can't see). Diagnostic only; never throws. - yield* Effect.sync(() => - LLMRequestPrep.recordCacheHitOutcome(ctx.sessionID, usage.tokens), - ).pipe(Effect.ignore) + yield* Effect.sync(() => LLMRequestPrep.recordCacheHitOutcome(ctx.sessionID, usage.tokens)).pipe( + Effect.ignore, + ) if (!ctx.assistantMessage.summary) { // TODO(v2): Temporary dual-write while migrating session messages to v2 events. if (mirrorAssistant) { @@ -1197,12 +1290,16 @@ export const layer = Layer.effect( const match = yield* readToolCall(toolCallID) if (!match) continue const part = match.part + const incompleteInput = part.state.status === "pending" + const toolError = incompleteInput + ? "Tool input was incomplete and was not executed" + : "Tool execution aborted" if (mirrorAssistant && match.call.assistantMessageID) { yield* events.publish(SessionEvent.Tool.Failed, { sessionID: ctx.sessionID, assistantMessageID: match.call.assistantMessageID, callID: toolCallID, - error: { type: "unknown", message: "Tool execution aborted" }, + error: { type: "unknown", message: toolError }, provider: { executed: part.metadata?.providerExecuted === true }, timestamp: DateTime.makeUnsafe(Date.now()), }) @@ -1214,8 +1311,8 @@ export const layer = Layer.effect( state: { ...part.state, status: "error", - error: "Tool execution aborted", - metadata: { ...metadata, interrupted: true }, + error: toolError, + metadata: { ...metadata, interrupted: true, incompleteInput }, time: { start: "time" in part.state ? part.state.time.start : end, end }, }, }) diff --git a/packages/deepagent-code/src/session/prompt.ts b/packages/deepagent-code/src/session/prompt.ts index 6d36fb83..313f5c04 100644 --- a/packages/deepagent-code/src/session/prompt.ts +++ b/packages/deepagent-code/src/session/prompt.ts @@ -61,10 +61,25 @@ import { Truncate } from "@/tool/truncate" import { Image } from "@/image/image" import { decodeDataUrl } from "@/util/data-url" import { Process } from "@/util/process" -import { Cause, Data, Effect, Exit, Latch, Layer, Option, Scope, Context, Schema, Types } from "effect" +import { + Cause, + Context, + Data, + Duration, + Effect, + Exit, + Fiber, + Latch, + Layer, + Option, + Schedule, + Schema, + Scope, + Types, +} from "effect" import * as EffectLogger from "@deepagent-code/core/effect/logger" import { InstanceState } from "@/effect/instance-state" -import { TaskTool, type TaskPromptOps } from "@/tool/task" +import { projectRecoveredSubagentRun, TaskTool, type TaskPromptOps } from "@/tool/task" import { SessionRunState } from "./run-state" import { SessionSteer } from "./steer" import { writeGovernanceAudit } from "./goal-governance-audit" @@ -94,6 +109,10 @@ import { LLMEvent } from "@deepagent-code/llm" import { ConversationLogWriter } from "./conversation-log-writer" import { collectVolatileFacts, refreshWorldState } from "./context-ledger" import { CodeIndexTrigger } from "./code-index-trigger" +import { ToolSemanticFingerprint } from "@/tool/semantic-fingerprint" +import { deliverTaskNotifications, recoverExpiredTaskRuns } from "@/tool/task-run" +import { registerDisposer, registerInitializer } from "@/effect/instance-registry" +import { InstanceRef } from "@/effect/instance-ref" // @ts-ignore globalThis.AI_SDK_LOG_WARNINGS = false @@ -139,6 +158,41 @@ function isOrphanedInterruptedTool(part: SessionV1.ToolPart) { return part.state.status === "error" && part.state.metadata?.interrupted === true } +function isStructuredFinalizer(metadata: unknown) { + if (!isRecord(metadata)) return false + if (!isRecord(metadata.deepagent)) return false + return isRecord(metadata.deepagent.structured_finalizer) +} + +function noninteractiveTaskActivity(metadata: unknown) { + if (!isRecord(metadata)) return false + if (!isRecord(metadata.deepagent)) return undefined + if (!isRecord(metadata.deepagent.task_activity)) return undefined + const activity = metadata.deepagent.task_activity + if (activity.interactive !== false) return undefined + if (!isRecord(activity.budget)) return { interactive: false as const } + const positive = (value: unknown) => + typeof value === "number" && Number.isSafeInteger(value) && value > 0 ? value : undefined + return { + interactive: false as const, + startedAt: positive(activity.started_at), + maxSteps: positive(activity.budget.max_steps), + maxTokens: positive(activity.budget.max_tokens), + maxWallMs: positive(activity.budget.max_wall_ms), + maxNoProgress: positive(activity.budget.max_no_progress), + } +} + +function taskNotification(metadata: unknown) { + if (!isRecord(metadata)) return undefined + if (!isRecord(metadata.deepagent)) return undefined + if (!isRecord(metadata.deepagent.task_notification)) return undefined + const runID = metadata.deepagent.task_notification.run_id + const outboxID = metadata.deepagent.task_notification.outbox_id + if (typeof runID !== "string" || typeof outboxID !== "string") return undefined + return { runID, outboxID } +} + // §S1.2 — a goal in one of these phases is no longer ticking, so a "goal_steer" would never be drained. // promptOrSteer routes to the plain "steer" channel (or a fresh turn) instead. Mirrors goal-manager's // isTerminalGoalPhase (kept as a local const to avoid a circular import: goal-manager imports this file). @@ -149,13 +203,9 @@ class InvalidInput extends Data.TaggedError("SessionPrompt.InvalidInput")<{ read // §S1.2 — convert PromptInput parts to the durable Prompt model used by the steer buffer. // All part types that have a Prompt equivalent are preserved; subtask parts are explicitly rejected // so they never produce a silent empty steer. The steer caller should surface this as a client error. -const promptInputToPrompt = ( - parts: PromptInput["parts"], -): Effect.Effect => { +const promptInputToPrompt = (parts: PromptInput["parts"]): Effect.Effect => { if (parts.some((p) => p.type === "subtask")) - return Effect.fail( - new InvalidInput({ message: "Subtask prompt parts cannot be steered while a session is busy" }), - ) + return Effect.fail(new InvalidInput({ message: "Subtask prompt parts cannot be steered while a session is busy" })) const text = parts .filter((p): p is Extract => p.type === "text") .map((p) => p.text) @@ -175,9 +225,7 @@ const promptInputToPrompt = ( .filter((p): p is Extract => p.type === "agent") .map((p) => new AgentAttachment({ name: p.name })) if (text.length === 0 && files.length === 0 && agents.length === 0) - return Effect.fail( - new InvalidInput({ message: "Steer prompt must contain at least one supported part" }), - ) + return Effect.fail(new InvalidInput({ message: "Steer prompt must contain at least one supported part" })) return Effect.succeed( Prompt.fromUserMessage({ text, @@ -1695,6 +1743,25 @@ export const layer = Layer.effect( const prompt: (input: PromptInput) => Effect.Effect = Effect.fn( "SessionPrompt.prompt", )(function* (input: PromptInput) { + const notification = taskNotification(input.metadata) + if (notification && input.messageID) { + const existing = yield* MessageV2.get({ sessionID: input.sessionID, messageID: input.messageID }).pipe( + Effect.provideService(Database.Service, database), + Effect.catchCause(() => Effect.succeed(undefined)), + ) + if (existing) { + const persisted = existing.info.role === "user" ? taskNotification(existing.info.metadata) : undefined + if ( + existing.info.role !== "user" || + persisted?.runID !== notification.runID || + persisted.outboxID !== notification.outboxID + ) + return yield* Effect.die( + new Error(`Task notification message ID ${input.messageID} conflicts with persisted content`), + ) + return existing + } + } const session = yield* sessions.get(input.sessionID).pipe(Effect.orDie) yield* revert.cleanup(session) const pipeline = yield* buildPromptPipelineSubmission(input) @@ -1722,6 +1789,7 @@ export const layer = Layer.effect( if (input.noReply === true) return message const first = yield* loop({ sessionID: input.sessionID }) + if (isStructuredFinalizer(input.metadata)) return first // V3 Plan A: mode-driven multi-round autonomous loop for high/max/ultra. It remains // fail-closed (any error -> the single-turn result). Real validation (A3), // git rollback (A5), revise turn, and the A3 macro-round suggestion are wired. @@ -1969,8 +2037,7 @@ export const layer = Layer.effect( ? { providerID: ProviderV2.ID.make(current.model.providerID), modelID: ModelV2.ID.make(current.model.id), - variant: - current.model.variant && current.model.variant !== "default" ? current.model.variant : undefined, + variant: current.model.variant && current.model.variant !== "default" ? current.model.variant : undefined, } : yield* currentModel(sessionID) const variant = "variant" in resolved ? resolved.variant : undefined @@ -2042,15 +2109,13 @@ export const layer = Layer.effect( return Option.getOrElse(decodeSoftLanding(raw), () => initialSoftLandingState) }) - const writeSoftLandingState: ( - sessionID: SessionID, - state: CompactionSoftLandingState, - ) => Effect.Effect = Effect.fn("SessionPrompt.writeSoftLandingState")(function* (sessionID, state) { - const session = yield* sessions.get(sessionID).pipe(Effect.orElseSucceed(() => undefined)) - // Merge into existing metadata so we never clobber a co-tenant key. - const metadata = { ...(session?.metadata ?? {}), [SOFT_LANDING_METADATA_KEY]: state } - yield* sessions.setMetadata({ sessionID, metadata }).pipe(Effect.ignore) - }) + const writeSoftLandingState: (sessionID: SessionID, state: CompactionSoftLandingState) => Effect.Effect = + Effect.fn("SessionPrompt.writeSoftLandingState")(function* (sessionID, state) { + const session = yield* sessions.get(sessionID).pipe(Effect.orElseSucceed(() => undefined)) + // Merge into existing metadata so we never clobber a co-tenant key. + const metadata = { ...(session?.metadata ?? {}), [SOFT_LANDING_METADATA_KEY]: state } + yield* sessions.setMetadata({ sessionID, metadata }).pipe(Effect.ignore) + }) // reminder (soft line): a lightweight, non-compacting tail nudge asking the model to persist key // decisions/findings into the plan's evidence/worklog. Reuses the SAME tail-user-message channel as @@ -2068,29 +2133,26 @@ export const layer = Layer.effect( text: string, model: { providerID: ProviderV2.ID; modelID: ModelV2.ID }, agentName: string, - ) => Effect.Effect = Effect.fn("SessionPrompt.injectTailReminder")(function* ( - sessionID, - text, - model, - agentName, - ) { - const msg = yield* sessions.updateMessage({ - id: MessageID.ascending(), - role: "user", - sessionID, - agent: agentName, - model, - time: { created: Date.now() }, - }) - yield* sessions.updatePart({ - id: PartID.ascending(), - messageID: msg.id, - sessionID, - type: "text", - synthetic: true, - text, - }) - }) + ) => Effect.Effect = Effect.fn("SessionPrompt.injectTailReminder")( + function* (sessionID, text, model, agentName) { + const msg = yield* sessions.updateMessage({ + id: MessageID.ascending(), + role: "user", + sessionID, + agent: agentName, + model, + time: { created: Date.now() }, + }) + yield* sessions.updatePart({ + id: PartID.ascending(), + messageID: msg.id, + sessionID, + type: "text", + synthetic: true, + text, + }) + }, + ) // fallback ("临终笔记" line): the last chance before a hard compaction. All tools stay available so // the model can call the plan-edit tool to固化 un-persisted state. Under a goal (loop/design) mode we @@ -2123,6 +2185,14 @@ export const layer = Layer.effect( "", ].join("\n") + const TOOL_INPUT_CONTINUE_TAIL_TEXT = [ + "", + "你上一轮的工具输入因达到输出长度上限而被截断，系统没有执行该工具，也没有应用其中的文件修改。", + "不要原样重发同一个大型 JSON 工具调用。将修改拆小；对于大型 write/edit/apply_patch，改用 `apply_patch_chunk`，", + "每个 patchText 块不超过 12000 UTF-8 字节（中文建议不超过约 4000 字）。begin 使用 offset 0；之后每次 append 和最终 commit 都使用上一结果返回的 nextOffset。", + "", + ].join("\n") + // V4.0.1 P1 (§3.3) — post-hard-compaction World State re-injection. After a hard compaction the // (now-narrowed) summary deliberately dropped file/env/diagnostics; this re-injects their LATEST // values as a TAIL user block (reuses the SAME injectTailReminder primitive — never the static system @@ -2134,17 +2204,14 @@ export const layer = Layer.effect( workspacePath: string | undefined, model: { providerID: ProviderV2.ID; modelID: ModelV2.ID }, agentName: string, - ) => Effect.Effect = Effect.fn("SessionPrompt.injectWorldStateTail")(function* ( - sessionID, - workspacePath, - model, - agentName, - ) { - if (!workspacePath) return - const facts = yield* collectVolatileFacts(workspacePath) - const rendered = yield* refreshWorldState({ workspacePath, facts }) - if (rendered.trim().length > 0) yield* injectTailReminder(sessionID, rendered, model, agentName) - }) + ) => Effect.Effect = Effect.fn("SessionPrompt.injectWorldStateTail")( + function* (sessionID, workspacePath, model, agentName) { + if (!workspacePath) return + const facts = yield* collectVolatileFacts(workspacePath) + const rendered = yield* refreshWorldState({ workspacePath, facts }) + if (rendered.trim().length > 0) yield* injectTailReminder(sessionID, rendered, model, agentName) + }, + ) const runLoop: (sessionID: SessionID, drainFirst?: boolean) => Effect.Effect = Effect.fn( "SessionPrompt.run", @@ -2183,7 +2250,30 @@ export const layer = Layer.effect( // V3.9 §A: `lsp` enables the AST symbol pass (symbol nodes + imports/calls edges) over the // content-sha-changed files; a language with no LSP client degrades to the file-level view. // SEAM: incremental mtime-gated fs-walking is the remaining follow-up (see code-index-trigger.ts). - if (!indexedSessions.has(sessionID)) { + const initialMessages = yield* MessageV2.filterCompactedEffect(sessionID).pipe( + Effect.provideService(Database.Service, database), + ) + const initialUser = MessageV2.latest(initialMessages).user + const initialFinalizer = isStructuredFinalizer(initialUser?.metadata) + const taskActivity = noninteractiveTaskActivity(initialUser?.metadata) || undefined + const failTaskBudget = Effect.fn("SessionPrompt.failTaskBudget")(function* ( + assistant: SessionV1.Assistant, + budget: "steps" | "tokens" | "wall_time", + limit: number, + used: number, + ) { + assistant.error = new SessionV1.TaskBudgetExceededError({ + message: `Subagent research ${budget} budget exhausted (${used}/${limit}).`, + budget, + limit, + used, + }).toObject() + assistant.finish = "error" + assistant.time.completed = Date.now() + yield* sessions.updateMessage(assistant) + yield* slog.warn("subagent.research.failed", { reason: "budget_exhausted", budget, limit, used }) + }) + if (!initialFinalizer && !indexedSessions.has(sessionID)) { indexedSessions.add(sessionID) yield* CodeIndexTrigger.indexWorkspace({ workspacePath: ctx.directory, fsys, lsp }).pipe( Effect.asVoid, @@ -2218,10 +2308,31 @@ export const layer = Layer.effect( const { user: lastUser, assistant: lastAssistant, finished: lastFinished, tasks } = MessageV2.latest(msgs) if (!lastUser) throw new Error("No user message found in stream. This should never happen.") + const finalizerMode = isStructuredFinalizer(lastUser.metadata) const lastAssistantMsg = msgs.findLast( (msg) => msg.info.role === "assistant" && msg.info.id === lastAssistant?.id, ) + const tokenUsage = taskActivity + ? msgs + .filter( + (item): item is SessionV1.WithParts & { info: SessionV1.Assistant } => + item.info.role === "assistant" && (!initialUser || item.info.id > initialUser.id), + ) + .reduce( + (sum, item) => sum + item.info.tokens.input + item.info.tokens.output + item.info.tokens.reasoning, + 0, + ) + : 0 + const elapsed = taskActivity?.startedAt ? Math.max(0, Date.now() - taskActivity.startedAt) : 0 + if (lastAssistant && taskActivity?.maxTokens && tokenUsage >= taskActivity.maxTokens) { + yield* failTaskBudget(lastAssistant, "tokens", taskActivity.maxTokens, tokenUsage) + break + } + if (lastAssistant && taskActivity?.maxWallMs && elapsed >= taskActivity.maxWallMs) { + yield* failTaskBudget(lastAssistant, "wall_time", taskActivity.maxWallMs, elapsed) + break + } // Some providers return "stop" even when the assistant message contains // tool calls. Keep the loop running so tool results can be sent back to // the model, but ignore cleanup-marked interrupted orphans. @@ -2229,6 +2340,9 @@ export const layer = Layer.effect( lastAssistantMsg?.parts.some( (part) => part.type === "tool" && !part.metadata?.providerExecuted && !isOrphanedInterruptedTool(part), ) ?? false + const orphan = lastAssistantMsg?.parts.find( + (part): part is SessionV1.ToolPart => part.type === "tool" && isOrphanedInterruptedTool(part), + ) if ( lastAssistant?.finish && @@ -2248,18 +2362,24 @@ export const layer = Layer.effect( const done = sls.outputContinuationCount ?? 0 if (done < outputContinuationMax()) { yield* writeSoftLandingState(sessionID, { ...sls, outputContinuationCount: done + 1 }) - yield* injectTailReminder(sessionID, OUTPUT_CONTINUE_TAIL_TEXT, lastUser.model, lastUser.agent) + yield* injectTailReminder( + sessionID, + orphan?.state.status === "error" && orphan.state.metadata?.incompleteInput === true + ? TOOL_INPUT_CONTINUE_TAIL_TEXT + : OUTPUT_CONTINUE_TAIL_TEXT, + lastUser.model, + lastUser.agent, + ) yield* slog.info("output soft-landing: continuing after length cutoff", { continuation: done + 1, max: outputContinuationMax(), }) continue } - yield* slog.warn("output soft-landing: continuation cap reached, ending turn", { max: outputContinuationMax() }) + yield* slog.warn("output soft-landing: continuation cap reached, ending turn", { + max: outputContinuationMax(), + }) } - const orphan = lastAssistantMsg?.parts.find( - (part): part is SessionV1.ToolPart => part.type === "tool" && isOrphanedInterruptedTool(part), - ) if (orphan) { yield* slog.warn("loop exit with orphaned interrupted tool", { messageID: lastAssistant.id, @@ -2272,14 +2392,18 @@ export const layer = Layer.effect( } // Output soft-landing: a natural stop (or any non-length finish that keeps looping via tool // calls) resets the consecutive-continuation run so a later length cutoff gets the full budget. - if (flags.outputSoftLanding && lastAssistant?.finish && lastAssistant.finish !== "length") { + if (!finalizerMode && flags.outputSoftLanding && lastAssistant?.finish && lastAssistant.finish !== "length") { const sls = yield* readSoftLandingState(sessionID) if ((sls.outputContinuationCount ?? 0) !== 0) yield* writeSoftLandingState(sessionID, { ...sls, outputContinuationCount: 0 }) } step++ - if (step === 1) { + if (lastAssistant && taskActivity?.maxSteps && step > taskActivity.maxSteps) { + yield* failTaskBudget(lastAssistant, "steps", taskActivity.maxSteps, step - 1) + break + } + if (step === 1 && !finalizerMode) { yield* title({ session, modelID: lastUser.model.modelID, @@ -2290,7 +2414,8 @@ export const layer = Layer.effect( } const model = yield* getModel(lastUser.model.providerID, lastUser.model.modelID, sessionID) - const task = tasks.pop() + const finalizerDecision = finalizerMode ? LLM.finalizerCapability(model) : undefined + const task = finalizerMode ? undefined : tasks.pop() if (task?.type === "subtask") { yield* handleSubtask({ task, model, lastUser, sessionID, session, msgs }) @@ -2315,7 +2440,7 @@ export const layer = Layer.effect( // "临终笔记" fallback (all tools retained), then the SAME hard compaction. `phase === "hard"` is // exactly `isOverflow`, and the reminder/fallback layers never move the hard line, so the // compaction trigger is unchanged. - if (lastFinished && lastFinished.summary !== true) { + if (!finalizerMode && lastFinished && lastFinished.summary !== true) { if (!flags.softLandingCompaction) { if (yield* compaction.isOverflow({ tokens: lastFinished.tokens, model })) { yield* compaction.create({ sessionID, agent: lastUser.agent, model: lastUser.model, auto: true }) @@ -2386,13 +2511,15 @@ export const layer = Layer.effect( yield* events.publish(Session.Event.Error, { sessionID, error: error.toObject() }) throw error } - const maxSteps = agent.steps ?? Infinity + const maxSteps = Math.min(agent.steps ?? Infinity, taskActivity?.maxSteps ?? Infinity) const isLastStep = step >= maxSteps - msgs = yield* SessionReminders.apply({ messages: msgs, agent, session }).pipe( - Effect.provideService(RuntimeFlags.Service, flags), - Effect.provideService(FSUtil.Service, fsys), - Effect.provideService(Session.Service, sessions), - ) + if (!finalizerMode) { + msgs = yield* SessionReminders.apply({ messages: msgs, agent, session }).pipe( + Effect.provideService(RuntimeFlags.Service, flags), + Effect.provideService(FSUtil.Service, fsys), + Effect.provideService(Session.Service, sessions), + ) + } const msg: SessionV1.Assistant = { id: MessageID.ascending(), @@ -2411,6 +2538,20 @@ export const layer = Layer.effect( } yield* sessions.updateMessage(msg) + if (finalizerDecision?.capability === "unsupported") { + msg.error = new NamedError.Unknown({ + message: `[${finalizerDecision.reason}] Structured finalization requires tool-call capability.`, + }).toObject() + msg.finish = "error" + msg.time.completed = Date.now() + yield* sessions.updateMessage(msg) + yield* slog.warn("subagent.finalize.failed", { + reason: finalizerDecision.reason, + protocol: finalizerDecision.protocol, + }) + break + } + const finalizeInterruptedAssistant = Effect.gen(function* () { if (msg.time.completed) return msg.error ??= MessageV2.fromError(new DOMException("Aborted", "AbortError"), { @@ -2427,6 +2568,8 @@ export const layer = Layer.effect( sessionID, model, sequenceTracker: toolSequenceTracker, + loopPolicy: finalizerMode || taskActivity ? "error" : "ask", + noProgressLimit: taskActivity?.maxNoProgress, }) .pipe(Effect.onInterrupt(() => finalizeInterruptedAssistant)) @@ -2435,22 +2578,24 @@ export const layer = Layer.effect( const bypassAgentCheck = lastUserMsg?.parts.some((p) => p.type === "agent") ?? false const promptOps = yield* ops() - const tools = yield* SessionTools.resolve({ - agent, - session, - model, - processor: handle, - bypassAgentCheck, - messages: msgs, - promptOps, - }).pipe( - Effect.provideService(Plugin.Service, plugin), - Effect.provideService(Permission.Service, permission), - Effect.provideService(ToolRegistry.Service, registry), - Effect.provideService(MCP.Service, mcp), - Effect.provideService(Truncate.Service, truncate), - Effect.provideService(RuntimeFlags.Service, flags), - ) + const tools: Record = finalizerMode + ? {} + : yield* SessionTools.resolve({ + agent, + session, + model, + processor: handle, + bypassAgentCheck, + messages: msgs, + promptOps, + }).pipe( + Effect.provideService(Plugin.Service, plugin), + Effect.provideService(Permission.Service, permission), + Effect.provideService(ToolRegistry.Service, registry), + Effect.provideService(MCP.Service, mcp), + Effect.provideService(Truncate.Service, truncate), + Effect.provideService(RuntimeFlags.Service, flags), + ) if (lastUser.format?.type === "json_schema") { tools["StructuredOutput"] = createStructuredOutputTool({ @@ -2460,8 +2605,14 @@ export const layer = Layer.effect( }, }) } + toolSequenceTracker.setFingerprintResolver((toolName, args) => + ToolSemanticFingerprint.resolve(tools[toolName], args), + ) + toolSequenceTracker.setResultFingerprintResolver((toolName, result) => + ToolSemanticFingerprint.resolveResult(tools[toolName], result), + ) - if (step === 1) + if (step === 1 && !finalizerMode) yield* summary.summarize({ sessionID, messageID: lastUser.id }).pipe(Effect.ignore, Effect.forkIn(scope)) if (step > 1 && lastFinished) { @@ -2482,7 +2633,7 @@ export const layer = Layer.effect( } } - yield* plugin.trigger("experimental.chat.messages.transform", {}, { messages: msgs }) + if (!finalizerMode) yield* plugin.trigger("experimental.chat.messages.transform", {}, { messages: msgs }) // PR-1: Compute terminal boundary for reasoning model-view projection. // The most recent settled assistant message (has finish, no pending tool calls) @@ -2503,18 +2654,29 @@ export const layer = Layer.effect( } } - const [skills, env, instructions, modelMsgs] = yield* Effect.all([ - sys.skills(agent), - sys.environment(model), - instruction.system().pipe(Effect.orDie), - MessageV2.toModelMessagesEffect(msgs, model, { terminalBoundaryID }), - ]) - const system = [...env, ...instructions, ...(skills ? [skills] : [])] const format = lastUser.format ?? { type: "text" as const } + const modelMsgs = yield* MessageV2.toModelMessagesEffect( + finalizerMode ? msgs.filter((item) => item.info.id === lastUser.id) : msgs, + model, + { terminalBoundaryID }, + ) + const system = finalizerMode + ? [ + buildStructuredOutputSystemPrompt(format.type === "json_schema" ? format.schema : {}), + "This is a bounded finalizer turn. Read the supplied research result and call StructuredOutput once. No research or other work is permitted.", + ] + : yield* Effect.all([ + sys.skills(agent), + sys.environment(model), + instruction.system().pipe(Effect.orDie), + ]).pipe( + Effect.map(([skills, env, instructions]) => [...env, ...instructions, ...(skills ? [skills] : [])]), + ) // P1: inject schema-aware prompt so the model knows the exact field names even // during extended-thinking (xhigh) reasoning where the tool definition may not // be immediately visible when the model starts generating its thinking tokens. - if (format.type === "json_schema") system.push(buildStructuredOutputSystemPrompt(format.schema)) + if (!finalizerMode && format.type === "json_schema") + system.push(buildStructuredOutputSystemPrompt(format.schema)) const result = yield* handle.process({ user: lastUser, agent, @@ -2525,7 +2687,8 @@ export const layer = Layer.effect( messages: [...modelMsgs, ...(isLastStep ? [{ role: "assistant" as const, content: MAX_STEPS }] : [])], tools, model, - toolChoice: format.type === "json_schema" ? "required" : undefined, + toolChoice: finalizerDecision?.toolChoice ?? (format.type === "json_schema" ? "required" : undefined), + reasoning: finalizerDecision?.reasoning, }) if (structured !== undefined) { @@ -2535,6 +2698,17 @@ export const layer = Layer.effect( return "break" as const } + if (finalizerMode) { + if (!handle.message.error) { + handle.message.error = new SessionV1.StructuredOutputError({ + message: "Finalizer did not produce valid structured output", + retries: 1, + }).toObject() + yield* sessions.updateMessage(handle.message) + } + return "break" as const + } + const finished = handle.message.finish && !["tool-calls", "unknown"].includes(handle.message.finish) if (finished && !handle.message.error) { if (format.type === "json_schema") { @@ -2567,9 +2741,8 @@ export const layer = Layer.effect( const currentAssistantMsg = latestMsgs.findLast( (m) => m.info.role === "assistant" && m.info.id === handle.message.id, ) - const hadStructuredOutputCall = currentAssistantMsg?.parts.some( - (p) => p.type === "tool" && p.tool === "StructuredOutput", - ) ?? false + const hadStructuredOutputCall = + currentAssistantMsg?.parts.some((p) => p.type === "tool" && p.tool === "StructuredOutput") ?? false if (hadStructuredOutputCall) { const retryMax = format.retryCount ?? 2 @@ -2637,7 +2810,7 @@ export const layer = Layer.effect( // it (codex: needsFollowUp = modelSaidContinue || pendingInput-nonempty). A non-consuming peek; // the actual drain (consume-once) happens at the next iteration's top. Gated by the flag. if (outcome === "break") { - if (flags.v4Steering && (yield* steerBuffer.hasPending(sessionID))) { + if (!finalizerMode && flags.v4Steering && (yield* steerBuffer.hasPending(sessionID))) { yield* slog.info("steer pending at model boundary, continuing to absorb") continue } @@ -2668,22 +2841,22 @@ export const layer = Layer.effect( prompt: Prompt delivery?: SessionSteer.Delivery messageID?: SessionMessage.ID - }) => Effect.Effect = Effect.fn( - "SessionPrompt.steer", - )(function* (input) { + }) => Effect.Effect = Effect.fn("SessionPrompt.steer")(function* (input) { if (!flags.v4Steering) return yield* Effect.die(new NamedError.Unknown({ message: "Steering is disabled (v4Steering=false)" })) const delivery = input.delivery ?? "steer" - const admitted = yield* steerBuffer.admit({ - sessionID: input.sessionID, - prompt: input.prompt, - delivery, - correlationID: input.messageID, - }).pipe( - Effect.catchTag("SessionSteer.CorrelationConflict", () => - Effect.die(new NamedError.Unknown({ message: "Steer correlation conflict: duplicate follow-up" })), - ), - ) + const admitted = yield* steerBuffer + .admit({ + sessionID: input.sessionID, + prompt: input.prompt, + delivery, + correlationID: input.messageID, + }) + .pipe( + Effect.catchTag("SessionSteer.CorrelationConflict", () => + Effect.die(new NamedError.Unknown({ message: "Steer correlation conflict: duplicate follow-up" })), + ), + ) yield* elog.info("steer admitted", { sessionID: input.sessionID, messageID: admitted.id, @@ -2721,9 +2894,7 @@ export const layer = Layer.effect( const goalActive = goal != null && !TERMINAL_GOAL_PHASES.has(goal.phase) if (goalActive) { const steerPrompt = yield* promptInputToPrompt(input.parts).pipe( - Effect.catchTag("SessionPrompt.InvalidInput", (e) => - Effect.die(e), - ), + Effect.catchTag("SessionPrompt.InvalidInput", (e) => Effect.die(e)), ) const admitted = yield* steer({ sessionID: input.sessionID, @@ -2746,9 +2917,7 @@ export const layer = Layer.effect( return { kind: "turn" as const, message } } const steerPrompt = yield* promptInputToPrompt(input.parts).pipe( - Effect.catchTag("SessionPrompt.InvalidInput", (e) => - Effect.die(e), - ), + Effect.catchTag("SessionPrompt.InvalidInput", (e) => Effect.die(e)), ) const admitted = yield* steer({ sessionID: input.sessionID, @@ -2895,6 +3064,71 @@ export const layer = Layer.effect( return result }) + const notificationWorkers = new Map>() + const startNotificationWorker = registerInitializer((ctx) => + Effect.runPromise( + Effect.gen(function* () { + if (notificationWorkers.has(ctx.directory)) return + const owner = `task-notification:${process.pid}:${randomUUID()}` + const pump = recoverExpiredTaskRuns({ directory: ctx.directory }).pipe( + Effect.tap((runs) => + Effect.forEach( + runs, + (run) => projectRecoveredSubagentRun(sessions, run).pipe(Effect.provideService(InstanceRef, ctx)), + { discard: true }, + ), + ), + Effect.flatMap(() => + deliverTaskNotifications({ + owner, + directory: ctx.directory, + limit: 1, + deliver: (item) => + prompt({ + messageID: item.messageID, + sessionID: item.parentSessionID, + agent: item.payload.agent, + variant: item.payload.variant, + metadata: { + deepagent: { + task_notification: { run_id: item.runID, outbox_id: item.id }, + }, + }, + parts: [{ type: "text", synthetic: true, text: item.payload.text }], + }).pipe(Effect.provideService(InstanceRef, ctx), Effect.asVoid), + }), + ), + Effect.provideService(Database.Service, database), + Effect.catchCause((cause) => + Effect.sync(() => + log.error("task notification pump failed", { directory: ctx.directory, cause: Cause.pretty(cause) }), + ).pipe(Effect.as([])), + ), + ) + const worker = yield* pump.pipe( + Effect.repeat(Schedule.spaced(Duration.seconds(2))), + Effect.asVoid, + Effect.forkIn(scope), + ) + notificationWorkers.set(ctx.directory, worker) + }), + ), + ) + const stopNotificationWorker = registerDisposer((directory) => { + const worker = notificationWorkers.get(directory) + if (!worker) return Promise.resolve() + notificationWorkers.delete(directory) + return Effect.runPromise(Fiber.interrupt(worker).pipe(Effect.asVoid)) + }) + yield* Effect.addFinalizer(() => + Effect.gen(function* () { + startNotificationWorker() + stopNotificationWorker() + yield* Effect.forEach(notificationWorkers.values(), Fiber.interrupt, { discard: true }) + notificationWorkers.clear() + }), + ) + return Service.of({ cancel, prompt, diff --git a/packages/deepagent-code/src/session/session.ts b/packages/deepagent-code/src/session/session.ts index 430d7b7f..58b51133 100644 --- a/packages/deepagent-code/src/session/session.ts +++ b/packages/deepagent-code/src/session/session.ts @@ -487,6 +487,7 @@ export interface Interface { readonly list: (input?: ListInput) => Effect.Effect readonly listGlobal: (input?: GlobalListInput) => Effect.Effect readonly create: (input?: { + id?: SessionID parentID?: SessionID title?: string agent?: string @@ -740,6 +741,7 @@ export const layer: Layer.Layer< }) const create = Effect.fn("Session.create")(function* (input?: { + id?: SessionID parentID?: SessionID title?: string agent?: string @@ -756,6 +758,7 @@ export const layer: Layer.Layer< const workspace = yield* InstanceState.workspaceID const directory = input?.directory ?? ctx.directory return yield* createNext({ + id: input?.id, parentID: input?.parentID, directory, path: sessionPath(ctx.worktree, directory), @@ -815,12 +818,10 @@ export const layer: Layer.Layer< input.isolate === "worktree" ? yield* Effect.serviceOption(Worktree.Service) : Option.none() const worktreeInfo = input.isolate === "worktree" && Option.isSome(worktreeOpt) - ? yield* worktreeOpt.value - .create({ name: `fork-${Identifier.ascending("session")}` }) - .pipe( - Effect.catchTag("WorktreeNotGitError", () => Effect.succeed(undefined)), - Effect.orDie, - ) + ? yield* worktreeOpt.value.create({ name: `fork-${Identifier.ascending("session")}` }).pipe( + Effect.catchTag("WorktreeNotGitError", () => Effect.succeed(undefined)), + Effect.orDie, + ) : undefined // 附-D 阶段3: resolve the effective fork directory. Precedence: a fresh worktree (阶段4) > diff --git a/packages/deepagent-code/src/session/tools.ts b/packages/deepagent-code/src/session/tools.ts index b4a4d56b..a6814a09 100644 --- a/packages/deepagent-code/src/session/tools.ts +++ b/packages/deepagent-code/src/session/tools.ts @@ -25,6 +25,7 @@ import { EffectBridge } from "@/effect/bridge" import { ProviderV2 } from "@deepagent-code/core/provider" import { ModelV2 } from "@deepagent-code/core/model" import { AgentGateway } from "@deepagent-code/core/agent-gateway" +import { ToolSemanticFingerprint } from "@/tool/semantic-fingerprint" const log = Log.create({ service: "session.tools" }) @@ -141,8 +142,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { type GateDirective = { kind: "block"; output: string } | { kind: "warn"; reason: string } | { kind: "pass" } const evaluatePlanGate = (sessionID: string, isMutating: boolean): GateDirective => { const latch = AgentGateway.DeepAgentSessionState.planLatch(sessionID) - const planStale = - latch?.latch === "stale" && !AgentGateway.DeepAgentPlanController.shouldEscapeToHuman(latch) + const planStale = latch?.latch === "stale" && !AgentGateway.DeepAgentPlanController.shouldEscapeToHuman(latch) // Gate strength must key off THIS session's EFFECTIVE mode, not the process-global one. The global // `snapshot().agentMode` ignores the per-request `agent_mode_override` (a downgraded subagent, or a // session pinned below the global) — so it would over- or under-gate a turn, and disagree with @@ -179,8 +179,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { : `${gateDecision.blockReason}. Call the \`plan\` tool first.` return { kind: "block", output } } - const gateWarnReason = - gateDecision.decision === "warn" && !lightweight ? gateDecision.blockReason : undefined + const gateWarnReason = gateDecision.decision === "warn" && !lightweight ? gateDecision.blockReason : undefined // A mutating tool that actually executes is forward progress → reset the consecutive-block counter. if (isMutating) { AgentGateway.DeepAgentSessionState.recordMutation(sessionID) @@ -214,7 +213,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { const command = (item.id === "bash" || item.id === "shell") && typeof (args as { command?: unknown } | undefined)?.command === "string" - ? ((args as { command: string }).command) + ? (args as { command: string }).command : null // Fail SAFE if the classifier ever throws (it is total today — pure regex/string ops — but a // future regex/refactor could introduce a throw): treat an unclassifiable command as mutating @@ -236,9 +235,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { // so this is a soft nudge: plan → stale but the next mutating tool still runs with a // reminder rather than being blocked. Effect.tapError(() => - Effect.sync(() => - AgentGateway.DeepAgentSessionState.markPlanStale(ctx.sessionID, "tool_failed"), - ), + Effect.sync(() => AgentGateway.DeepAgentSessionState.markPlanStale(ctx.sessionID, "tool_failed")), ), ) const withReminder = @@ -270,6 +267,8 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { // M2 (S1-v3.4): carry the registry's explicit provenance onto the freshly // built AI SDK tool so request.ts reads it instead of guessing from the name. if (item.provenance) ToolProvenance.set(aiToolDef, item.provenance) + if (item.semanticFingerprint) ToolSemanticFingerprint.set(aiToolDef, item.semanticFingerprint) + if (item.resultFingerprint) ToolSemanticFingerprint.setResult(aiToolDef, item.resultFingerprint) tools[item.id] = aiToolDef } diff --git a/packages/deepagent-code/src/tool/semantic-fingerprint.ts b/packages/deepagent-code/src/tool/semantic-fingerprint.ts new file mode 100644 index 00000000..b1527774 --- /dev/null +++ b/packages/deepagent-code/src/tool/semantic-fingerprint.ts @@ -0,0 +1,26 @@ +import type { Tool } from "ai" + +export type Resolver = (input: unknown) => unknown + +const resolvers = new WeakMap() +const resultResolvers = new WeakMap() + +export function set(tool: Tool, resolver: Resolver) { + resolvers.set(tool, resolver) +} + +export function resolve(tool: Tool | undefined, input: unknown) { + const resolver = tool && resolvers.get(tool) + return resolver ? resolver(input) : input +} + +export function setResult(tool: Tool, resolver: (result: Result) => unknown) { + resultResolvers.set(tool, (result) => resolver(result as Result)) +} + +export function resolveResult(tool: Tool | undefined, result: unknown) { + const resolver = tool && resultResolvers.get(tool) + return resolver?.(result) +} + +export * as ToolSemanticFingerprint from "./semantic-fingerprint" diff --git a/packages/deepagent-code/src/tool/shell.ts b/packages/deepagent-code/src/tool/shell.ts index 618443c8..23b4684c 100644 --- a/packages/deepagent-code/src/tool/shell.ts +++ b/packages/deepagent-code/src/tool/shell.ts @@ -629,6 +629,20 @@ export const ShellTool = Tool.define( return { description: prompt.description, parameters: prompt.parameters, + semanticFingerprint: (params: Parameters) => ({ + command: params.command, + workdir: params.workdir ?? null, + timeout: params.timeout ?? defaultTimeoutMs, + }), + resultFingerprint: (result) => ({ + exit: result.metadata.exit, + output: new Bun.CryptoHasher("sha256").update(result.output).digest("hex"), + attachments: result.attachments?.map((attachment) => ({ + mime: attachment.mime, + filename: attachment.filename, + url: attachment.url.startsWith("data:") ? "data" : attachment.url, + })), + }), execute: (params: Parameters, ctx: Tool.Context) => Effect.gen(function* () { const instanceCtx = yield* InstanceState.context diff --git a/packages/deepagent-code/src/tool/task-run.ts b/packages/deepagent-code/src/tool/task-run.ts new file mode 100644 index 00000000..70c5dab5 --- /dev/null +++ b/packages/deepagent-code/src/tool/task-run.ts @@ -0,0 +1,897 @@ +import { Database } from "@deepagent-code/core/database/database" +import { + TaskAdmissionTable, + TaskNotificationOutboxTable, + TaskRunTable, + SessionTable, +} from "@deepagent-code/core/session/sql" +import { and, asc, eq, gt, inArray, isNull, lte, max, or, sql } from "drizzle-orm" +import { Cause, Data, Effect } from "effect" +import { Identifier } from "@/id/id" +import { MessageID, SessionID } from "@/session/schema" + +export type State = + | "admitted" + | "provisioning" + | "researching" + | "finalizing" + | "completed" + | "error" + | "cancelled" + | "interrupted" +export type Phase = "admission" | "research" | "finalize" | "settled" +export type DeliveryMode = "foreground" | "background" +export type ErrorData = { code: string; message: string; data?: Record } +export type NotificationPayload = { agent: string; variant?: string; text: string } + +export type Run = { + runID: string + rootRunID?: string + requestHash: string + parentSessionID: SessionID + parentMessageID: MessageID + toolCallID: string + childSessionID: SessionID + generation: number + deliveryMode: DeliveryMode + phase: Phase + state: State + reason?: string + attempts: number + executionOwner?: string + leaseExpiresAt?: number + rawResultMessageID?: MessageID + structuredResultMessageID?: MessageID + output?: string + error?: ErrorData + timeCreated: number + timeUpdated: number + timeSettled?: number +} + +export type Admission = { + run: Run + exactRetry: boolean + runCreated: boolean +} + +export type OutboxItem = { + id: string + runID: string + messageID: MessageID + parentSessionID: SessionID + directory: string + payload: NotificationPayload + attempts: number +} + +export class AdmissionConflict extends Data.TaggedError("TaskRun.AdmissionConflict")<{ + readonly admissionKey: string + readonly reason: "request" | "delivery" | "child" | "join" +}> {} + +class ConcurrentAdmission extends Data.TaggedError("TaskRun.ConcurrentAdmission")<{ + readonly admissionKey: string +}> {} + +const terminalStates: ReadonlyArray = ["completed", "error", "cancelled", "interrupted"] +const activeStates: ReadonlyArray = ["admitted", "provisioning", "researching", "finalizing"] + +const canonicalJson = (value: unknown): string => { + if (value === null) return "null" + if (value === undefined) return '"__undefined__"' + if (Array.isArray(value)) return `[${value.map(canonicalJson).join(",")}]` + if (typeof value !== "object") return JSON.stringify(value) ?? "null" + return `{${Object.entries(value) + .sort(([a], [b]) => a.localeCompare(b)) + .map(([key, item]) => `${JSON.stringify(key)}:${canonicalJson(item)}`) + .join(",")}}` +} + +export const requestHash = (value: unknown) => new Bun.CryptoHasher("sha256").update(canonicalJson(value)).digest("hex") + +const fromRow = (row: typeof TaskRunTable.$inferSelect): Run => ({ + runID: row.run_id, + rootRunID: row.root_run_id ?? undefined, + requestHash: row.request_hash, + parentSessionID: SessionID.make(row.parent_session_id), + parentMessageID: MessageID.ascending(row.parent_message_id), + toolCallID: row.tool_call_id, + childSessionID: SessionID.make(row.child_session_id), + generation: row.generation, + deliveryMode: row.delivery_mode, + phase: row.phase, + state: row.state, + reason: row.reason ?? undefined, + attempts: row.attempts, + executionOwner: row.execution_owner ?? undefined, + leaseExpiresAt: row.lease_expires_at ?? undefined, + rawResultMessageID: row.raw_result_message_id ?? undefined, + structuredResultMessageID: row.structured_result_message_id ?? undefined, + output: row.output ?? undefined, + error: row.error ?? undefined, + timeCreated: row.time_created, + timeUpdated: row.time_updated, + timeSettled: row.time_settled ?? undefined, +}) + +const admissionKey = (input: { parentSessionID: SessionID; parentMessageID: MessageID; toolCallID: string }) => + `${input.parentSessionID}\u0000${input.parentMessageID}\u0000${input.toolCallID}` + +export function admitTaskRun(input: { + parentSessionID: SessionID + parentMessageID: MessageID + toolCallID: string + childSessionID?: SessionID + joinRunID?: string + request: unknown + deliveryMode: DeliveryMode + now?: number +}) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const key = admissionKey(input) + const hash = requestHash(input.request) + const childSessionID = input.childSessionID ?? SessionID.create() + const now = input.now ?? Date.now() + const transact = () => + db.transaction((tx) => + Effect.gen(function* () { + const existingAdmission = yield* tx + .select() + .from(TaskAdmissionTable) + .where(eq(TaskAdmissionTable.admission_key, key)) + .get() + .pipe(Effect.orDie) + if (existingAdmission) { + if (existingAdmission.request_hash !== hash) + return yield* Effect.fail(new AdmissionConflict({ admissionKey: key, reason: "request" })) + if (existingAdmission.delivery_mode !== input.deliveryMode) + return yield* Effect.fail(new AdmissionConflict({ admissionKey: key, reason: "delivery" })) + const run = yield* tx + .select() + .from(TaskRunTable) + .where(eq(TaskRunTable.run_id, existingAdmission.run_id)) + .get() + .pipe(Effect.orDie) + if (!run) return yield* Effect.die(`Task admission ${key} references a missing run`) + if (input.childSessionID !== undefined && run.child_session_id !== input.childSessionID) + return yield* Effect.fail(new AdmissionConflict({ admissionKey: key, reason: "child" })) + return { run: fromRow(run), exactRetry: true, runCreated: false } satisfies Admission + } + + const joined = input.joinRunID + ? yield* tx + .select() + .from(TaskRunTable) + .where( + and( + eq(TaskRunTable.run_id, input.joinRunID), + eq(TaskRunTable.child_session_id, childSessionID), + inArray(TaskRunTable.state, ["researching", "finalizing"]), + ), + ) + .get() + .pipe(Effect.orDie) + : undefined + if (input.joinRunID && !joined) + return yield* Effect.fail(new AdmissionConflict({ admissionKey: key, reason: "join" })) + + const conflictingActive = + !joined && input.childSessionID + ? yield* tx + .select({ run_id: TaskRunTable.run_id }) + .from(TaskRunTable) + .where( + and( + eq(TaskRunTable.child_session_id, input.childSessionID), + inArray(TaskRunTable.state, [...activeStates]), + ), + ) + .get() + .pipe(Effect.orDie) + : undefined + if (conflictingActive) return yield* Effect.fail(new AdmissionConflict({ admissionKey: key, reason: "join" })) + + const insertedRun = joined + ? undefined + : yield* Effect.gen(function* () { + for (let retry = 0; retry < 4; retry++) { + const latest = yield* tx + .select({ generation: max(TaskRunTable.generation) }) + .from(TaskRunTable) + .where(eq(TaskRunTable.child_session_id, childSessionID)) + .get() + .pipe(Effect.orDie) + const runID = Identifier.ascending("job") + const inserted = yield* tx + .insert(TaskRunTable) + .values({ + run_id: runID, + root_run_id: runID, + request_hash: hash, + parent_session_id: input.parentSessionID, + parent_message_id: input.parentMessageID, + tool_call_id: input.toolCallID, + child_session_id: childSessionID, + generation: (latest?.generation ?? 0) + 1, + delivery_mode: input.deliveryMode, + phase: "admission", + state: "admitted", + time_created: now, + time_updated: now, + }) + .onConflictDoNothing() + .returning() + .get() + .pipe(Effect.orDie) + if (inserted) return inserted + } + return yield* Effect.die("TaskRun.admit could not allocate a unique child generation") + }) + const run = joined ?? insertedRun + if (!run) return yield* Effect.die("TaskRun.admit did not resolve a run") + const insertedAdmission = yield* tx + .insert(TaskAdmissionTable) + .values({ + admission_key: key, + request_hash: hash, + run_id: run.run_id, + parent_session_id: input.parentSessionID, + parent_message_id: input.parentMessageID, + tool_call_id: input.toolCallID, + delivery_mode: input.deliveryMode, + time_created: now, + }) + .onConflictDoNothing() + .returning() + .get() + .pipe(Effect.orDie) + if (!insertedAdmission) return yield* Effect.fail(new ConcurrentAdmission({ admissionKey: key })) + return { run: fromRow(run), exactRetry: false, runCreated: insertedRun !== undefined } satisfies Admission + }), + ) + const attempt = (remaining: number): Effect.Effect => + transact().pipe( + Effect.catchTag("SqlError", Effect.die), + Effect.catchTag("TaskRun.ConcurrentAdmission", () => + remaining > 0 + ? attempt(remaining - 1) + : Effect.die(`Task admission ${key} remained contended after bounded retries`), + ), + ) + return yield* attempt(4) + }) +} + +export function spawnTaskTakeover(input: { root: Run; childSessionID: SessionID; now?: number }) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + return yield* db.transaction((tx) => + Effect.gen(function* () { + const latest = yield* tx + .select({ generation: max(TaskRunTable.generation) }) + .from(TaskRunTable) + .where(eq(TaskRunTable.child_session_id, input.childSessionID)) + .get() + .pipe(Effect.orDie) + const runID = Identifier.ascending("job") + const inserted = yield* tx + .insert(TaskRunTable) + .values({ + run_id: runID, + root_run_id: input.root.rootRunID ?? input.root.runID, + request_hash: input.root.requestHash, + parent_session_id: input.root.parentSessionID, + parent_message_id: input.root.parentMessageID, + tool_call_id: `${input.root.toolCallID}:takeover:${runID}`, + child_session_id: input.childSessionID, + generation: (latest?.generation ?? 0) + 1, + delivery_mode: input.root.deliveryMode, + phase: "admission", + state: "admitted", + time_created: now, + time_updated: now, + }) + .returning() + .get() + .pipe(Effect.orDie) + return fromRow(inserted) + }), + ) + }) +} + +export function claimTaskProvisioning(input: { run: Run; owner: string; now?: number; leaseMs?: number }) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const updated = yield* db + .update(TaskRunTable) + .set({ + state: "provisioning", + execution_owner: input.owner, + lease_expires_at: now + (input.leaseMs ?? 30_000), + time_updated: now, + }) + .where( + and( + eq(TaskRunTable.run_id, input.run.runID), + eq(TaskRunTable.generation, input.run.generation), + or( + eq(TaskRunTable.state, "admitted"), + and( + eq(TaskRunTable.state, "provisioning"), + or(eq(TaskRunTable.execution_owner, input.owner), lte(TaskRunTable.lease_expires_at, now)), + ), + ), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + return updated ? fromRow(updated) : undefined + }) +} + +export function startTaskRun(run: Run, owner: string, now = Date.now(), leaseMs = 30_000) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const updated = yield* db + .update(TaskRunTable) + .set({ + phase: "research", + state: "researching", + execution_owner: owner, + lease_expires_at: now + leaseMs, + time_updated: now, + }) + .where( + and( + eq(TaskRunTable.run_id, run.runID), + eq(TaskRunTable.generation, run.generation), + eq(TaskRunTable.state, "provisioning"), + eq(TaskRunTable.execution_owner, owner), + gt(TaskRunTable.lease_expires_at, now), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + return updated ? fromRow(updated) : undefined + }) +} + +export function renewTaskRunLease(input: { run: Run; owner: string; now?: number; leaseMs?: number }) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const updated = yield* db + .update(TaskRunTable) + .set({ + lease_expires_at: now + (input.leaseMs ?? 30_000), + time_updated: now, + }) + .where( + and( + eq(TaskRunTable.run_id, input.run.runID), + eq(TaskRunTable.generation, input.run.generation), + eq(TaskRunTable.execution_owner, input.owner), + inArray(TaskRunTable.state, ["provisioning", "researching", "finalizing"]), + gt(TaskRunTable.lease_expires_at, now), + ), + ) + .returning({ run_id: TaskRunTable.run_id }) + .get() + .pipe(Effect.orDie) + return updated !== undefined + }) +} + +export function recoverExpiredTaskRuns(input: { directory: string; now?: number; nullLeaseGraceMs?: number }) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const nullLeaseBefore = now - (input.nullLeaseGraceMs ?? 30_000) + return yield* Effect.uninterruptible( + db.transaction((tx) => + Effect.gen(function* () { + const candidates = yield* tx + .select({ run: TaskRunTable, agent: SessionTable.agent }) + .from(TaskRunTable) + .innerJoin(SessionTable, eq(SessionTable.id, TaskRunTable.parent_session_id)) + .where( + and( + eq(SessionTable.directory, input.directory), + inArray(TaskRunTable.state, ["provisioning", "researching", "finalizing"]), + or( + lte(TaskRunTable.lease_expires_at, now), + and(isNull(TaskRunTable.lease_expires_at), lte(TaskRunTable.time_updated, nullLeaseBefore)), + ), + ), + ) + .all() + .pipe(Effect.orDie) + return yield* Effect.forEach( + candidates, + (candidate) => + Effect.gen(function* () { + const updated = yield* tx + .update(TaskRunTable) + .set({ + phase: "settled", + state: "error", + reason: "execution_lease_expired", + error: { + code: "execution_lease_expired", + message: "The task execution lease expired before the run reached a durable terminal state.", + }, + execution_owner: null, + lease_expires_at: null, + time_updated: now, + time_settled: now, + }) + .where( + and( + eq(TaskRunTable.run_id, candidate.run.run_id), + eq(TaskRunTable.generation, candidate.run.generation), + inArray(TaskRunTable.state, ["provisioning", "researching", "finalizing"]), + or( + lte(TaskRunTable.lease_expires_at, now), + and(isNull(TaskRunTable.lease_expires_at), lte(TaskRunTable.time_updated, nullLeaseBefore)), + ), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + if (!updated) return undefined + yield* tx + .insert(TaskNotificationOutboxTable) + .values({ + id: `task-notify:${updated.run_id}`, + run_id: updated.run_id, + message_id: MessageID.ascending(), + parent_session_id: updated.parent_session_id, + directory: input.directory, + payload: { + agent: candidate.agent ?? "build", + text: + candidate.run.state === "provisioning" + ? [ + "A subagent stopped while its durable child session was being provisioned.", + "No provider work will be resumed automatically. Retry with a new task request; the original tool call remains terminal.", + ].join("\n") + : [ + "A subagent stopped because its execution lease expired before a durable result was recorded.", + `Partial work is preserved; call task_read({ task_id: "${updated.child_session_id}" }) before retrying.`, + ].join("\n"), + }, + status: "pending", + attempts: 0, + available_at: now, + time_created: now, + time_updated: now, + }) + .onConflictDoNothing() + .run() + .pipe(Effect.orDie) + return fromRow(updated) + }), + { concurrency: 1 }, + ).pipe(Effect.map((runs) => runs.filter((run): run is Run => run !== undefined))) + }), + ), + ) + }) +} + +export function markTaskResearchCompleted(run: Run, owner: string, rawResultMessageID: MessageID, now = Date.now()) { + return updateActive( + run, + owner, + { raw_result_message_id: rawResultMessageID, time_updated: now }, + ["researching"], + now, + ) +} + +export function markTaskFinalizing( + run: Run, + owner: string, + attempt: number, + rawResultMessageID: MessageID, + now = Date.now(), +) { + return updateActive( + run, + owner, + { + phase: "finalize", + state: "finalizing", + attempts: attempt, + raw_result_message_id: rawResultMessageID, + time_updated: now, + }, + ["researching", "finalizing"], + now, + ) +} + +export function markTaskFinalized(run: Run, owner: string, structuredResultMessageID: MessageID, now = Date.now()) { + return updateActive( + run, + owner, + { structured_result_message_id: structuredResultMessageID, time_updated: now }, + ["finalizing"], + now, + ) +} + +function updateActive( + run: Run, + owner: string, + values: Partial, + states: State[], + now: number, +) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const updated = yield* db + .update(TaskRunTable) + .set(values) + .where( + and( + eq(TaskRunTable.run_id, run.runID), + eq(TaskRunTable.generation, run.generation), + eq(TaskRunTable.execution_owner, owner), + inArray(TaskRunTable.state, states), + gt(TaskRunTable.lease_expires_at, now), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + return updated ? fromRow(updated) : undefined + }) +} + +export function settleTaskRun(input: { + run: Run + owner: string + state: Extract + reason: string + output?: string + error?: ErrorData + structuredResultMessageID?: MessageID + notification?: { directory: string; payload: NotificationPayload } + now?: number +}) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + return yield* Effect.uninterruptible( + db.transaction((tx) => + Effect.gen(function* () { + const updated = yield* tx + .update(TaskRunTable) + .set({ + phase: "settled", + state: input.state, + reason: input.reason, + output: input.output, + error: input.error, + structured_result_message_id: input.structuredResultMessageID, + execution_owner: null, + lease_expires_at: null, + time_updated: now, + time_settled: now, + }) + .where( + and( + eq(TaskRunTable.run_id, input.run.runID), + eq(TaskRunTable.generation, input.run.generation), + eq(TaskRunTable.execution_owner, input.owner), + inArray(TaskRunTable.state, [...activeStates]), + gt(TaskRunTable.lease_expires_at, now), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + if (!updated) { + const existing = yield* tx + .select() + .from(TaskRunTable) + .where(eq(TaskRunTable.run_id, input.run.runID)) + .get() + .pipe(Effect.orDie) + return { won: false as const, run: existing ? fromRow(existing) : input.run } + } + if (input.notification) { + yield* tx + .insert(TaskNotificationOutboxTable) + .values({ + id: `task-notify:${input.run.runID}`, + run_id: input.run.runID, + message_id: MessageID.ascending(), + parent_session_id: input.run.parentSessionID, + directory: input.notification.directory, + payload: input.notification.payload, + status: "pending", + attempts: 0, + available_at: now, + time_created: now, + time_updated: now, + }) + .onConflictDoNothing() + .run() + .pipe(Effect.orDie) + } + return { won: true as const, run: fromRow(updated) } + }), + ), + ) + }) +} + +export function getTaskRunByAdmission(input: { + parentSessionID: SessionID + parentMessageID: MessageID + toolCallID: string +}) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const row = yield* db + .select({ run: TaskRunTable }) + .from(TaskAdmissionTable) + .innerJoin(TaskRunTable, eq(TaskRunTable.run_id, TaskAdmissionTable.run_id)) + .where(eq(TaskAdmissionTable.admission_key, admissionKey(input))) + .get() + .pipe(Effect.orDie) + return row ? fromRow(row.run) : undefined + }) +} + +export function getActiveTaskRunByChild(childSessionID: SessionID) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const row = yield* db + .select() + .from(TaskRunTable) + .where(and(eq(TaskRunTable.child_session_id, childSessionID), inArray(TaskRunTable.state, [...activeStates]))) + .get() + .pipe(Effect.orDie) + return row ? fromRow(row) : undefined + }) +} + +export function getTaskRun(runID: string) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const row = yield* db.select().from(TaskRunTable).where(eq(TaskRunTable.run_id, runID)).get().pipe(Effect.orDie) + return row ? fromRow(row) : undefined + }) +} + +export function claimTaskNotifications(input: { + owner: string + directory: string + now?: number + leaseMs?: number + limit?: number +}) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const leaseUntil = now + (input.leaseMs ?? 30_000) + return yield* db.transaction((tx) => + Effect.gen(function* () { + const candidates = yield* tx + .select({ id: TaskNotificationOutboxTable.id }) + .from(TaskNotificationOutboxTable) + .where( + and( + eq(TaskNotificationOutboxTable.directory, input.directory), + lte(TaskNotificationOutboxTable.available_at, now), + or( + eq(TaskNotificationOutboxTable.status, "pending"), + and( + eq(TaskNotificationOutboxTable.status, "delivering"), + or( + isNull(TaskNotificationOutboxTable.lease_expires_at), + lte(TaskNotificationOutboxTable.lease_expires_at, now), + ), + ), + ), + ), + ) + .orderBy(asc(TaskNotificationOutboxTable.available_at), asc(TaskNotificationOutboxTable.id)) + .limit(input.limit ?? 20) + .all() + .pipe(Effect.orDie) + const claimed: OutboxItem[] = [] + for (const candidate of candidates) { + const row = yield* tx + .update(TaskNotificationOutboxTable) + .set({ + status: "delivering", + lease_owner: input.owner, + lease_expires_at: leaseUntil, + attempts: sql`${TaskNotificationOutboxTable.attempts} + 1`, + time_updated: now, + }) + .where( + and( + eq(TaskNotificationOutboxTable.id, candidate.id), + eq(TaskNotificationOutboxTable.directory, input.directory), + or( + eq(TaskNotificationOutboxTable.status, "pending"), + and( + eq(TaskNotificationOutboxTable.status, "delivering"), + or( + isNull(TaskNotificationOutboxTable.lease_expires_at), + lte(TaskNotificationOutboxTable.lease_expires_at, now), + ), + ), + ), + ), + ) + .returning() + .get() + .pipe(Effect.orDie) + if (!row) continue + claimed.push({ + id: row.id, + runID: row.run_id, + messageID: MessageID.ascending(row.message_id), + parentSessionID: SessionID.make(row.parent_session_id), + directory: row.directory, + payload: row.payload, + attempts: row.attempts, + }) + } + return claimed + }), + ) + }) +} + +export function acknowledgeTaskNotification(input: { id: string; owner: string; attempts: number; now?: number }) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const updated = yield* db + .update(TaskNotificationOutboxTable) + .set({ + status: "delivered", + lease_owner: null, + lease_expires_at: null, + last_error: null, + time_updated: now, + time_delivered: now, + }) + .where( + and( + eq(TaskNotificationOutboxTable.id, input.id), + eq(TaskNotificationOutboxTable.status, "delivering"), + eq(TaskNotificationOutboxTable.lease_owner, input.owner), + eq(TaskNotificationOutboxTable.attempts, input.attempts), + ), + ) + .returning({ id: TaskNotificationOutboxTable.id }) + .get() + .pipe(Effect.orDie) + return updated !== undefined + }) +} + +export function rejectTaskNotification(input: { + id: string + owner: string + attempts: number + error: string + now?: number + maxAttempts?: number +}) { + return Effect.gen(function* () { + const { db } = yield* Database.Service + const now = input.now ?? Date.now() + const dead = input.attempts >= (input.maxAttempts ?? 10) + const delay = Math.min(300_000, 1_000 * 2 ** Math.max(0, input.attempts - 1)) + const updated = yield* db + .update(TaskNotificationOutboxTable) + .set({ + status: dead ? "dead" : "pending", + available_at: dead ? now : now + delay, + lease_owner: null, + lease_expires_at: null, + last_error: input.error.slice(0, 4_000), + time_updated: now, + }) + .where( + and( + eq(TaskNotificationOutboxTable.id, input.id), + eq(TaskNotificationOutboxTable.status, "delivering"), + eq(TaskNotificationOutboxTable.lease_owner, input.owner), + eq(TaskNotificationOutboxTable.attempts, input.attempts), + ), + ) + .returning({ id: TaskNotificationOutboxTable.id }) + .get() + .pipe(Effect.orDie) + return updated ? dead : undefined + }) +} + +export function deliverTaskNotifications(input: { + owner: string + directory: string + deliver: (item: OutboxItem) => Effect.Effect + now?: () => number + leaseMs?: number + limit?: number + maxAttempts?: number +}) { + return Effect.gen(function* () { + const items = yield* claimTaskNotifications({ + owner: input.owner, + directory: input.directory, + now: input.now?.(), + leaseMs: input.leaseMs, + limit: input.limit, + }) + return yield* Effect.forEach( + items, + (item) => + input.deliver(item).pipe( + Effect.flatMap(() => + acknowledgeTaskNotification({ + id: item.id, + owner: input.owner, + attempts: item.attempts, + now: input.now?.(), + }).pipe( + Effect.tap((acknowledged) => + Effect.logInfo("subagent.notification.delivered").pipe( + Effect.annotateLogs({ + run_id: item.runID, + outbox_id: item.id, + parent_session_id: item.parentSessionID, + acknowledged, + }), + ), + ), + ), + ), + Effect.catchCause((cause) => + rejectTaskNotification({ + id: item.id, + owner: input.owner, + attempts: item.attempts, + error: Cause.pretty(cause), + now: input.now?.(), + maxAttempts: input.maxAttempts, + }).pipe( + Effect.tap((dead) => + Effect.logWarning("subagent.notification.failed").pipe( + Effect.annotateLogs({ + run_id: item.runID, + outbox_id: item.id, + parent_session_id: item.parentSessionID, + attempts: item.attempts, + dead: dead === true, + }), + ), + ), + Effect.as(false), + ), + ), + ), + { concurrency: 1 }, + ) + }) +} + +export const isTerminal = (run: Run) => terminalStates.includes(run.state) diff --git a/packages/deepagent-code/src/tool/task.ts b/packages/deepagent-code/src/tool/task.ts index 5269ae6d..2986c611 100644 --- a/packages/deepagent-code/src/tool/task.ts +++ b/packages/deepagent-code/src/tool/task.ts @@ -20,7 +20,7 @@ import { import { evaluate as evaluatePermission } from "../permission" import type { SessionPrompt } from "../session/prompt" import { Config } from "@/config/config" -import { Cause, Effect, Exit, Option, Schema, Scope } from "effect" +import { Cause, Duration, Effect, Exit, Option, Schedule, Schema, Scope } from "effect" import { EffectBridge } from "@/effect/bridge" import { RuntimeFlags } from "@/effect/runtime-flags" import { Database } from "@deepagent-code/core/database/database" @@ -29,8 +29,32 @@ import { Git } from "@/git" import { Orchestration } from "../agent/schema/orchestration" import { Orchestration as CoreOrchestration } from "@deepagent-code/core/deepagent/orchestration" import { AgentGateway } from "@deepagent-code/core/agent-gateway" -import { downgradeOneLevel } from "@deepagent-code/core/deepagent/mode" +import { downgradeOneLevel, type AgentMode } from "@deepagent-code/core/deepagent/mode" +import { ModelV2 } from "@deepagent-code/core/model" +import { ProviderV2 } from "@deepagent-code/core/provider" import { TaskConcurrency } from "./task-concurrency" +import Ajv from "ajv" +import { KeyedMutex } from "@deepagent-code/core/effect/keyed-mutex" +import { Log } from "@deepagent-code/core/util/log" +import { + admitTaskRun, + claimTaskProvisioning, + deliverTaskNotifications, + getActiveTaskRunByChild, + getTaskRun, + isTerminal, + markTaskFinalized, + markTaskFinalizing, + markTaskResearchCompleted, + renewTaskRunLease, + settleTaskRun, + spawnTaskTakeover, + startTaskRun, + type ErrorData, + type Run as DurableTaskRun, +} from "./task-run" + +const taskLog = Log.create({ service: "tool.task" }) /** * L3 (v3.8.0 §L3): resolve the task tool's optional `output_schema` param into a raw JSON Schema @@ -73,6 +97,565 @@ export function resolveOutputSchema( return ToolJsonSchema.fromSchema(schema) as unknown as Record } +const FINALIZER_ATTEMPTS = 2 +const FINALIZER_RAW_RESULT_MAX_CHARS = 80_000 +export const DEFAULT_SUBAGENT_RESEARCH_BUDGET = { + maxSteps: 64, + maxTokens: 200_000, + maxWallMs: 30 * 60_000, + maxNoProgress: 6, +} as const + +export type SubagentResearchBudget = { + readonly maxSteps: number + readonly maxTokens: number + readonly maxWallMs: number + readonly maxNoProgress: number +} + +export type SubagentPromptInput = { + ops: TaskPromptOps + prompt: string + sessionID: SessionID + model: { modelID: ModelV2.ID; providerID: ProviderV2.ID } + variant: string | undefined + agent: string + agentModeOverride: AgentMode | undefined + outputSchema: Record | undefined + runID?: string + budget?: SubagentResearchBudget + tools: Record + worktreeInfo: Worktree.Info | undefined + onResearchCompleted?: (sourceMessageID: MessageID) => Effect.Effect + onFinalizing?: (input: { attempt: number; sourceMessageID: MessageID }) => Effect.Effect + onFinalized?: (messageID: MessageID) => Effect.Effect +} + +type SubagentTerminalReason = + | "structured_output_valid" + | "text_output_valid" + | "provider_error" + | "structured_output_missing" + | "structured_output_invalid" + | "doom_loop" + | "assistant_error" + | "human" + | "parent_interrupted" + | "timeout" + | "takeover" + | "budget_exhausted" + | "execution_lease_expired" + | "runtime_error" +const subagentSettlementLocks = KeyedMutex.makeUnsafe() + +function isRecord(value: unknown): value is Record { + return typeof value === "object" && value !== null && !Array.isArray(value) +} + +function subagentMetadata(metadata: Session.Info["metadata"]) { + const deepagent = isRecord(metadata?.deepagent) ? metadata.deepagent : {} + return { + deepagent, + subagent: isRecord(deepagent.subagent) ? deepagent.subagent : {}, + } +} + +function stringField(value: unknown, key: string) { + if (!isRecord(value)) return undefined + return typeof value[key] === "string" ? value[key] : undefined +} + +function terminalReason(error: string | undefined): SubagentTerminalReason { + const code = error?.match(/^\[([^\]]+)\]/)?.[1] + if (code === "provider_error") return code + if (code === "structured_output_missing") return code + if (code === "structured_output_invalid") return code + if (code === "doom_loop") return code + if (code === "assistant_error") return code + if (code === "budget_exhausted") return code + return "runtime_error" +} + +function projectSubagentRun(sessions: Session.Interface, run: DurableTaskRun, continueActive = false) { + const sessionID = run.childSessionID + return subagentSettlementLocks.withLock(sessionID)( + Effect.gen(function* () { + const current = yield* sessions.get(sessionID).pipe(Effect.orDie) + const { deepagent, subagent: previous } = subagentMetadata(current.metadata) + if ( + continueActive && + previous.finished !== true && + typeof previous.run_id === "string" && + typeof previous.generation === "number" + ) { + return { runID: previous.run_id, generation: previous.generation } + } + yield* sessions.setMetadata({ + sessionID, + metadata: { + ...current.metadata, + deepagent: { + ...deepagent, + subagent: { + finished: false, + state: "researching", + phase: "research", + run_id: run.runID, + generation: run.generation, + attempts: 0, + started_at: Date.now(), + }, + }, + }, + }) + return run + }), + ) +} + +function lostTaskRunLease(run: DurableTaskRun) { + return new Error(`Task run ${run.runID} lost its execution lease or terminal settlement race`) +} + +function markSubagentResearchCompleted(run: DurableTaskRun, owner: string, sourceMessageID: MessageID) { + return markTaskResearchCompleted(run, owner, sourceMessageID).pipe( + Effect.flatMap((updated) => (updated ? Effect.void : Effect.fail(lostTaskRunLease(run)))), + ) +} + +function markSubagentFinalized(run: DurableTaskRun, owner: string, messageID: MessageID) { + return markTaskFinalized(run, owner, messageID).pipe( + Effect.flatMap((updated) => (updated ? Effect.void : Effect.fail(lostTaskRunLease(run)))), + ) +} + +function markSubagentFinalizing( + sessions: Session.Interface, + run: DurableTaskRun, + owner: string, + input: { attempt: number; sourceMessageID: MessageID }, +) { + const sessionID = run.childSessionID + return subagentSettlementLocks.withLock(sessionID)( + Effect.gen(function* () { + const updated = yield* markTaskFinalizing(run, owner, input.attempt, input.sourceMessageID) + if (!updated) yield* Effect.fail(lostTaskRunLease(run)) + const current = yield* sessions.get(sessionID).pipe(Effect.orDie) + const { deepagent, subagent } = subagentMetadata(current.metadata) + if (subagent.run_id !== run.runID || subagent.generation !== run.generation || subagent.finished === true) return + yield* sessions.setMetadata({ + sessionID, + metadata: { + ...current.metadata, + deepagent: { + ...deepagent, + subagent: { + ...subagent, + state: "finalizing", + phase: "finalize", + attempts: input.attempt, + raw_result_ref: input.sourceMessageID, + }, + }, + }, + }) + }), + ) +} + +function settleSubagentRun( + sessions: Session.Interface, + run: DurableTaskRun, + owner: string, + state: "completed" | "error" | "cancelled" | "interrupted", + reason: SubagentTerminalReason, + input?: { + output?: string + error?: ErrorData + structuredResultMessageID?: MessageID + notification?: { directory: string; agent: string; variant?: string; text: string } + }, +) { + const sessionID = run.childSessionID + return subagentSettlementLocks.withLock(sessionID)( + Effect.gen(function* () { + const settled = yield* settleTaskRun({ + run, + owner, + state, + reason, + output: input?.output, + error: input?.error, + structuredResultMessageID: input?.structuredResultMessageID, + notification: input?.notification + ? { + directory: input.notification.directory, + payload: { + agent: input.notification.agent, + variant: input.notification.variant, + text: input.notification.text, + }, + } + : undefined, + }) + if (!settled.won) return false + const current = yield* sessions.get(sessionID).pipe(Effect.orDie) + const { deepagent, subagent } = subagentMetadata(current.metadata) + if (subagent.run_id !== run.runID || subagent.generation !== run.generation || subagent.finished === true) + return true + yield* sessions.setMetadata({ + sessionID, + metadata: { + ...current.metadata, + deepagent: { + ...deepagent, + subagent: { + ...subagent, + finished: true, + state, + phase: "settled", + settled_at: Date.now(), + reason, + }, + }, + }, + }) + taskLog.info("subagent.settled", { + run_id: run.runID, + child_session_id: sessionID, + state, + reason, + parent_notification: input?.notification !== undefined, + }) + return settled.won + }), + ) +} + +export function projectRecoveredSubagentRun(sessions: Session.Interface, run: DurableTaskRun) { + const sessionID = run.childSessionID + return subagentSettlementLocks.withLock(sessionID)( + Effect.gen(function* () { + const current = yield* sessions.get(sessionID).pipe(Effect.catchCause(() => Effect.succeed(undefined))) + if (!current) return + const { deepagent, subagent } = subagentMetadata(current.metadata) + if (subagent.run_id !== run.runID || subagent.generation !== run.generation || subagent.finished === true) return + yield* sessions.setMetadata({ + sessionID, + metadata: { + ...current.metadata, + deepagent: { + ...deepagent, + subagent: { + ...subagent, + finished: true, + state: "error", + phase: "settled", + settled_at: run.timeSettled ?? Date.now(), + reason: "execution_lease_expired", + }, + }, + }, + }) + }), + ) +} + +function withTaskRunLease(run: DurableTaskRun, owner: string, effect: Effect.Effect) { + const heartbeat = renewTaskRunLease({ run, owner }).pipe( + Effect.flatMap((renewed) => + renewed ? Effect.void : Effect.fail(new Error(`Task run ${run.runID} lost its execution lease`)), + ), + Effect.repeat(Schedule.spaced(Duration.seconds(10))), + Effect.flatMap(() => Effect.never), + ) + return Effect.raceFirst(effect, heartbeat) +} + +function taskError(input: { + code: string + message: string + sessionID: SessionID + phase: "research" | "finalize" + attempts?: number +}) { + return new Error( + `[${input.code}] ${input.message} ` + + `Child session: ${input.sessionID}. Phase: ${input.phase}.` + + (input.attempts === undefined ? "" : ` Attempts: ${input.attempts}.`) + + ` Partial work is preserved; call task_read({ task_id: "${input.sessionID}" }) before retrying.`, + ) +} + +function assistantError(result: SessionV1.WithParts): { name: string; message: string } | undefined { + if (result.info.role !== "assistant" || !result.info.error) return undefined + return { + name: result.info.error.name ?? "UnknownError", + message: stringField(result.info.error.data, "message") ?? result.info.error.name ?? "Assistant failed", + } +} + +function assistantText(result: SessionV1.WithParts) { + return result.parts + .filter((item): item is SessionV1.TextPart => item.type === "text" && !item.synthetic && !item.ignored) + .map((item) => item.text) + .join("\n") + .trim() +} + +function validateStructuredOutput(schema: Record, value: unknown): string | undefined { + const { $schema: _, ...document } = schema + const validate = new Ajv({ allErrors: true, strict: false }).compile(document) + if (validate(value)) return undefined + return ( + validate.errors?.map((error) => `${error.instancePath || "/"} ${error.message ?? "is invalid"}`).join("; ") ?? + "schema validation failed" + ) +} + +export function runSubagentPrompt(input: SubagentPromptInput): Effect.Effect { + return Effect.gen(function* () { + const parts = yield* input.ops.resolvePromptParts(input.prompt) + const startedAt = Date.now() + const budget = input.budget ?? DEFAULT_SUBAGENT_RESEARCH_BUDGET + taskLog.info("subagent.research.started", { + run_id: input.runID, + child_session_id: input.sessionID, + max_steps: budget.maxSteps, + max_tokens: budget.maxTokens, + max_wall_ms: budget.maxWallMs, + }) + const research = yield* input.ops + .prompt({ + messageID: MessageID.ascending(), + sessionID: input.sessionID, + model: input.model, + variant: input.variant, + agent: input.agent, + metadata: { + deepagent: { + task_activity: { + interactive: false, + run_id: input.runID, + started_at: startedAt, + budget: { + max_steps: budget.maxSteps, + max_tokens: budget.maxTokens, + max_wall_ms: budget.maxWallMs, + max_no_progress: budget.maxNoProgress, + }, + }, + ...(input.agentModeOverride ? { agent_mode_override: input.agentModeOverride } : {}), + }, + }, + tools: input.tools, + parts: input.worktreeInfo + ? [ + { + type: "text" as const, + text: + `You are running in an ISOLATED git worktree at ${input.worktreeInfo.directory} (branch ${input.worktreeInfo.branch ?? "detached"}). ` + + `You inherited context from the parent session, but your working directory is this worktree. ` + + `Re-read files before editing (do not trust remembered paths/contents), and know your changes stay isolated until merged back.`, + }, + ...parts, + ] + : parts, + }) + .pipe( + Effect.timeout(Duration.millis(budget.maxWallMs)), + Effect.catchIf(Cause.isTimeoutError, () => + Effect.sync(() => { + taskLog.warn("subagent.research.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + reason: "wall_time_budget_exhausted", + }) + }).pipe( + Effect.andThen( + Effect.fail( + taskError({ + code: "budget_exhausted", + message: `Research wall-time budget exhausted (${budget.maxWallMs}ms).`, + sessionID: input.sessionID, + phase: "research", + }), + ), + ), + ), + ), + ) + const researchError = assistantError(research) + if (researchError) { + taskLog.warn("subagent.research.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + reason: researchError.name, + }) + return yield* Effect.fail( + taskError({ + code: + researchError.name === "APIError" + ? "provider_error" + : researchError.name === "DoomLoopError" + ? "doom_loop" + : researchError.name === "TaskBudgetExceededError" + ? "budget_exhausted" + : "assistant_error", + message: `${researchError.name}: ${researchError.message}`, + sessionID: input.sessionID, + phase: "research", + }), + ) + } + taskLog.info("subagent.research.completed", { + run_id: input.runID, + child_session_id: input.sessionID, + result_message_id: research.info.id, + }) + if (input.onResearchCompleted) yield* input.onResearchCompleted(research.info.id) + if (!input.outputSchema) return research.parts.findLast((item) => item.type === "text")?.text ?? "" + const raw = assistantText(research) + if (!raw) { + taskLog.warn("subagent.research.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + reason: "research_output_missing", + }) + return yield* Effect.fail( + taskError({ + code: "research_output_missing", + message: "Research completed without a textual result to finalize.", + sessionID: input.sessionID, + phase: "research", + }), + ) + } + + const boundedRaw = Array.from(raw).slice(0, FINALIZER_RAW_RESULT_MAX_CHARS).join("") + let correction: string | undefined + taskLog.info("subagent.finalize.started", { + run_id: input.runID, + child_session_id: input.sessionID, + max_attempts: FINALIZER_ATTEMPTS, + }) + for (let attempt = 1; attempt <= FINALIZER_ATTEMPTS; attempt++) { + taskLog.info("subagent.finalize.attempted", { + run_id: input.runID, + child_session_id: input.sessionID, + attempt, + }) + if (input.onFinalizing) yield* input.onFinalizing({ attempt, sourceMessageID: research.info.id }) + const finalized = yield* input.ops.prompt({ + messageID: MessageID.ascending(), + sessionID: input.sessionID, + model: input.model, + variant: input.variant, + agent: input.agent, + format: new SessionV1.OutputFormatJsonSchema({ + type: "json_schema", + schema: input.outputSchema, + retryCount: 1, + }), + metadata: { + deepagent: { + ...(input.agentModeOverride ? { agent_mode_override: input.agentModeOverride } : {}), + structured_finalizer: { + attempt, + source_message_id: research.info.id, + }, + }, + }, + parts: [ + { + type: "text", + text: [ + "Convert the persisted research result below into the requested StructuredOutput schema.", + "Do not continue research and do not add facts that are absent from the result.", + correction ? `Previous validation error: ${correction}` : "", + "", + boundedRaw, + "", + ] + .filter(Boolean) + .join("\n"), + }, + ], + }) + const error = assistantError(finalized) + if (error) { + correction = `${error.name}: ${error.message}`.slice(0, 1_000) + if (error.name === "StructuredOutputError" && attempt < FINALIZER_ATTEMPTS) continue + taskLog.warn("subagent.finalize.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + attempt, + reason: error.name, + }) + return yield* Effect.fail( + taskError({ + code: error.name === "APIError" ? "provider_error" : "structured_output_invalid", + message: correction, + sessionID: input.sessionID, + phase: "finalize", + attempts: attempt, + }), + ) + } + const structured = finalized.info.role === "assistant" ? finalized.info.structured : undefined + if (structured === undefined) { + correction = "Model did not call StructuredOutput." + if (attempt < FINALIZER_ATTEMPTS) continue + taskLog.warn("subagent.finalize.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + attempt, + reason: "structured_output_missing", + }) + return yield* Effect.fail( + taskError({ + code: "structured_output_missing", + message: correction, + sessionID: input.sessionID, + phase: "finalize", + attempts: attempt, + }), + ) + } + const validationError = validateStructuredOutput(input.outputSchema, structured) + if (!validationError) { + if (input.onFinalized) yield* input.onFinalized(finalized.info.id) + taskLog.info("subagent.finalize.completed", { + run_id: input.runID, + child_session_id: input.sessionID, + attempt, + result_message_id: finalized.info.id, + }) + return JSON.stringify(structured) + } + const boundedValidationError = validationError.slice(0, 1_000) + correction = boundedValidationError + if (attempt < FINALIZER_ATTEMPTS) continue + taskLog.warn("subagent.finalize.failed", { + run_id: input.runID, + child_session_id: input.sessionID, + attempt, + reason: "structured_output_invalid", + }) + return yield* Effect.fail( + taskError({ + code: "structured_output_invalid", + message: boundedValidationError, + sessionID: input.sessionID, + phase: "finalize", + attempts: attempt, + }), + ) + } + return yield* Effect.die(new Error("unreachable: structured finalizer exhausted without settlement")) + }) +} + export interface TaskPromptOps { cancel(sessionID: SessionID): Effect.Effect resolvePromptParts(template: string): Effect.Effect @@ -178,12 +761,22 @@ type AttemptMetadata = Record & { readonly background?: boolean } +type SettlementDetails = { + readonly output?: string + readonly error?: ErrorData + readonly notifyText?: string +} + interface AttemptBundle { readonly worktreeInfo: Worktree.Info | undefined readonly worktree: Worktree.Interface | undefined readonly nextSession: Session.Info readonly metadata: AttemptMetadata - readonly markFinished: (state: "completed" | "error" | "cancelled" | "interrupted", reason?: "human" | "parent_interrupted" | "timeout" | "takeover" | "runtime_error") => Effect.Effect + readonly markFinished: ( + state: "completed" | "error" | "cancelled" | "interrupted", + reason?: SubagentTerminalReason, + details?: SettlementDetails, + ) => Effect.Effect readonly inject: (state: "completed" | "error", text: string, takeovers: number) => Effect.Effect readonly automaticWriteIsolation: boolean readonly mergeWorktree: () => Effect.Effect @@ -240,9 +833,9 @@ export const TaskTool = Tool.define( ? yield* sessions.get(SessionID.make(params.task_id)).pipe(Effect.catchCause(() => Effect.succeed(undefined))) : undefined const parent = yield* sessions.get(ctx.sessionID) - const parentAgent = parent.agent - ? yield* agent.get(parent.agent).pipe(Effect.catchCause(() => Effect.succeed(undefined))) - : undefined + const parentAgent = yield* agent + .get(parent.agent ?? ctx.agent) + .pipe(Effect.catchCause(() => Effect.succeed(undefined))) // F5: unified child admission — resolves depth once and validates for ALL creation paths // (takeover and default). Resume (task_id present) takes the validation-only branch; a fresh @@ -292,6 +885,168 @@ export const TaskTool = Tool.define( // childDepth is used by BOTH the takeover path and the default path when writing metadata. const childDepth = parentDepth + 1 + // Runtime tool calls always carry callID. Direct programmatic callers (primarily tests and + // embedded integrations) predate that contract, so give those invocations a unique identity; + // exact-retry semantics are available only when the caller supplies the stable callID. + const toolCallID = ctx.callID ?? Identifier.ascending("tool") + const msg = yield* MessageV2.get({ sessionID: ctx.sessionID, messageID: ctx.messageID }).pipe( + Effect.provideService(Database.Service, database), + Effect.orDie, + ) + if (msg.info.role !== "assistant") return yield* Effect.fail(new Error("Not an assistant message")) + const variant = msg.info.variant + const model = next.model ?? { + modelID: msg.info.modelID, + providerID: msg.info.providerID, + } + const ops = ctx.extra?.promptOps as TaskPromptOps + if (!ops) return yield* Effect.fail(new Error("TaskTool requires promptOps in ctx.extra")) + const resolvedOutputSchema = resolveOutputSchema(params.output_schema, params.subagent_type) + const subagentIntensity = + (cfg.provider?.deepagent?.options?.subagentIntensity as string | undefined) === "downgrade" + ? "downgrade" + : "inherit" + const childAgentModeOverride = + subagentIntensity === "downgrade" ? downgradeOneLevel(AgentGateway.snapshot().agentMode) : undefined + const caps: CoreOrchestration.OrchestrationCaps = { + maxFanout: cfg.experimental?.orchestration?.max_fanout, + maxConcurrency: cfg.experimental?.orchestration?.max_concurrency, + } + const agentMaxConcurrency = next.limits?.maxConcurrency + const researchBudget: SubagentResearchBudget = { + maxSteps: flags.subagentResearchStepLimit ?? DEFAULT_SUBAGENT_RESEARCH_BUDGET.maxSteps, + maxTokens: flags.subagentResearchTokenLimit ?? DEFAULT_SUBAGENT_RESEARCH_BUDGET.maxTokens, + maxWallMs: flags.subagentResearchWallMs ?? DEFAULT_SUBAGENT_RESEARCH_BUDGET.maxWallMs, + maxNoProgress: flags.subagentNoProgressLimit ?? DEFAULT_SUBAGENT_RESEARCH_BUDGET.maxNoProgress, + } + const activeRun = session + ? yield* getActiveTaskRunByChild(session.id).pipe(Effect.provideService(Database.Service, database)) + : undefined + const activeJob = activeRun ? yield* background.get(activeRun.childSessionID) : undefined + const admission = yield* admitTaskRun({ + parentSessionID: ctx.sessionID, + parentMessageID: ctx.messageID, + toolCallID, + childSessionID: session?.id, + joinRunID: activeRun?.runID, + request: params, + deliveryMode: runInBackground ? "background" : "foreground", + }).pipe(Effect.provideService(Database.Service, database)) + const executionOwner = + activeJob?.status === "running" && + activeRun?.executionOwner !== undefined && + activeRun.leaseExpiresAt !== undefined && + activeRun.leaseExpiresAt > Date.now() + ? activeRun.executionOwner + : `${process.pid}:${Identifier.ascending("job")}` + if (admission.exactRetry && isTerminal(admission.run)) { + if (admission.run.state === "completed") { + return { + title: params.description, + metadata: { + parentSessionId: ctx.sessionID, + sessionId: admission.run.childSessionID, + subagentType: params.subagent_type, + model, + }, + output: renderOutput({ + sessionID: admission.run.childSessionID, + state: "completed", + summary: "Task result replayed from durable settlement", + text: admission.run.output ?? "", + maxChars: flags.subagentOutputMaxChars, + }), + } + } + return yield* Effect.fail( + new Error( + admission.run.error?.message ?? + `Task ${admission.run.childSessionID} previously settled as ${admission.run.state} (${admission.run.reason ?? "unknown"})`, + ), + ) + } + const shouldProvision = + admission.runCreated || (admission.exactRetry && ["admitted", "provisioning"].includes(admission.run.state)) + const claimedRun = shouldProvision + ? yield* claimTaskProvisioning({ run: admission.run, owner: executionOwner }).pipe( + Effect.provideService(Database.Service, database), + ) + : undefined + if (admission.exactRetry && !claimedRun) { + return { + title: params.description, + metadata: { + parentSessionId: ctx.sessionID, + sessionId: admission.run.childSessionID, + subagentType: params.subagent_type, + model, + background: true, + jobId: admission.run.childSessionID, + }, + output: renderOutput({ + sessionID: admission.run.childSessionID, + state: "running", + summary: "Task admission replayed without duplicate execution", + text: BACKGROUND_UPDATED, + maxChars: flags.subagentOutputMaxChars, + }), + } + } + if (shouldProvision && !claimedRun) + return yield* Effect.fail(new Error(`Task run ${admission.run.runID} lost its provisioning claim`)) + const admittedSession = + session ?? + (yield* sessions.get(admission.run.childSessionID).pipe(Effect.catchCause(() => Effect.succeed(undefined)))) + const ownsActiveRun = (run: DurableTaskRun) => + (run.state === "researching" || run.state === "finalizing") && + run.executionOwner === executionOwner && + run.leaseExpiresAt !== undefined && + run.leaseExpiresAt > Date.now() + const activateRun = (run: DurableTaskRun) => + ownsActiveRun(run) + ? projectSubagentRun(sessions, run, true).pipe(Effect.as(run)) + : startTaskRun(run, executionOwner).pipe( + Effect.provideService(Database.Service, database), + Effect.flatMap((started) => + started + ? projectSubagentRun(sessions, started).pipe(Effect.as(started)) + : getTaskRun(run.runID).pipe( + Effect.provideService(Database.Service, database), + Effect.flatMap((current) => + current && ownsActiveRun(current) + ? projectSubagentRun(sessions, current, true).pipe(Effect.as(current)) + : Effect.fail(new Error(`Task run ${run.runID} could not enter researching state`)), + ), + ), + ), + ) + const notification = (text: string) => ({ + directory: parent.directory, + agent: parent.agent ?? ctx.agent, + variant, + text, + }) + const dispatchNotifications = () => + deliverTaskNotifications({ + owner: `${executionOwner}:notification`, + directory: parent.directory, + deliver: (item) => + ops + .prompt({ + messageID: item.messageID, + sessionID: item.parentSessionID, + agent: item.payload.agent, + variant: item.payload.variant, + metadata: { + deepagent: { + task_notification: { run_id: item.runID, outbox_id: item.id }, + }, + }, + parts: [{ type: "text", synthetic: true, text: item.payload.text }], + }) + .pipe(Effect.asVoid), + }).pipe(Effect.provideService(Database.Service, database), Effect.asVoid) + // v4.0.4 块1 (1a+1b): timeout + takeover — enabled ONLY when DEEPAGENT_CODE_SUBAGENT_TIMEOUT_MS // is set; when it is not, execution falls through to the default path below, which stays // byte-identical to the pre-flag behavior. timeout and takeover are an inseparable unit: a bare @@ -303,35 +1058,10 @@ export const TaskTool = Tool.define( const timeoutMs = flags.subagentTimeoutMs const takeoverLimit = flags.subagentTakeoverLimit ?? 2 - const msg = yield* MessageV2.get({ sessionID: ctx.sessionID, messageID: ctx.messageID }).pipe( - Effect.provideService(Database.Service, database), - Effect.orDie, - ) - if (msg.info.role !== "assistant") return yield* Effect.fail(new Error("Not an assistant message")) - const variant = msg.info.variant - const model = next.model ?? { - modelID: msg.info.modelID, - providerID: msg.info.providerID, - } - const ops = ctx.extra?.promptOps as TaskPromptOps - if (!ops) return yield* Effect.fail(new Error("TaskTool requires promptOps in ctx.extra")) - const resolvedOutputSchema = resolveOutputSchema(params.output_schema, params.subagent_type) - const subagentIntensity = - (cfg.provider?.deepagent?.options?.subagentIntensity as string | undefined) === "downgrade" - ? "downgrade" - : "inherit" - const childAgentModeOverride = - subagentIntensity === "downgrade" ? downgradeOneLevel(AgentGateway.snapshot().agentMode) : undefined - const caps: CoreOrchestration.OrchestrationCaps = { - maxFanout: cfg.experimental?.orchestration?.max_fanout, - maxConcurrency: cfg.experimental?.orchestration?.max_concurrency, - } - const agentMaxConcurrency = next.limits?.maxConcurrency - // A fresh attempt gets its own worktree (same fork base as the discarded one) and a brand-new // child session; the resumed session (task_id) is only reused by the FIRST attempt. - const spawnAttempt = Effect.fn("TaskTool.spawnAttempt")(function* (first: boolean) { - const resumed = first ? session : undefined + const spawnAttempt = Effect.fn("TaskTool.spawnAttempt")(function* (first: boolean, runState: DurableTaskRun) { + const resumed = first ? admittedSession : undefined const isolate = !resumed && (params.isolation === "worktree" || subagentIsWriteType(next)) const worktreeOpt = isolate ? yield* Effect.serviceOption(Worktree.Service) @@ -345,6 +1075,7 @@ export const TaskTool = Tool.define( const nextSession = resumed ?? (yield* sessions.create({ + id: runState.childSessionID, parentID: ctx.sessionID, title: params.description + ` (@${next.name} subagent)`, agent: next.name, @@ -371,10 +1102,17 @@ export const TaskTool = Tool.define( }) const startAttempt = Effect.fn("TaskTool.startAttempt")(function* ( - a: { worktree: Worktree.Interface | undefined; worktreeInfo: Worktree.Info | undefined; nextSession: Session.Info }, + a: { + worktree: Worktree.Interface | undefined + worktreeInfo: Worktree.Info | undefined + nextSession: Session.Info + }, + runState: DurableTaskRun, takeovers: number, allowExtend: boolean, ) { + const activeRunState = yield* activateRun(runState) + const activeOwner = activeRunState.executionOwner ?? executionOwner const metadata = { parentSessionId: ctx.sessionID, sessionId: a.nextSession.id, @@ -392,121 +1130,82 @@ export const TaskTool = Tool.define( }) const runTaskInner = Effect.fn("TaskTool.runTaskInner")(function* () { - const parts = yield* ops.resolvePromptParts(params.prompt) - const promptParts = a.worktreeInfo - ? [ - { - type: "text" as const, - text: - `You are running in an ISOLATED git worktree at ${a.worktreeInfo.directory} (branch ${a.worktreeInfo.branch ?? "detached"}). ` + - `You inherited context from the parent session, but your working directory is this worktree. ` + - `Re-read files before editing (do not trust remembered paths/contents), and know your changes stay isolated until merged back.`, - }, - ...parts, - ] - : parts - const result = yield* ops.prompt({ - messageID: MessageID.ascending(), + return yield* runSubagentPrompt({ + ops, + prompt: params.prompt, sessionID: a.nextSession.id, - model: { - modelID: model.modelID, - providerID: model.providerID, - }, + model, variant: next.model ? undefined : variant, agent: next.name, - ...(childAgentModeOverride - ? { metadata: { deepagent: { agent_mode_override: childAgentModeOverride } } } - : {}), - ...(resolvedOutputSchema - ? { - format: new SessionV1.OutputFormatJsonSchema({ - type: "json_schema", - schema: resolvedOutputSchema, - }), - } - : {}), + agentModeOverride: childAgentModeOverride, + outputSchema: resolvedOutputSchema, + runID: activeRunState.runID, + budget: researchBudget, + worktreeInfo: a.worktreeInfo, + onResearchCompleted: (messageID) => + markSubagentResearchCompleted(activeRunState, activeOwner, messageID).pipe( + Effect.provideService(Database.Service, database), + ), + onFinalizing: (progress) => + markSubagentFinalizing(sessions, activeRunState, activeOwner, progress).pipe( + Effect.provideService(Database.Service, database), + ), + onFinalized: (messageID) => + markSubagentFinalized(activeRunState, activeOwner, messageID).pipe( + Effect.provideService(Database.Service, database), + ), tools: { - // F5: use evaluatePermission (not presence check) — "ask" is not sufficient, only - // an explicit "allow" rule on the subagent's own permission makes these tools visible. - ...(evaluatePermission("todowrite", "*", next.permission).action === "allow" ? {} : { todowrite: false }), - // task_status inspects the PARENT session's dispatched-subagent list, so it is a - // task-management capability: gate it on the same `task` permission. A subagent that - // cannot dispatch tasks has no sibling list to inspect, and (like `task`) must not be - // able to reach into task orchestration unless explicitly granted. - ...(evaluatePermission(id, "*", next.permission).action === "allow" ? {} : { task: false, task_status: false }), + ...(evaluatePermission("todowrite", "*", next.permission).action === "allow" + ? {} + : { todowrite: false }), + ...(evaluatePermission(id, "*", next.permission).action === "allow" + ? {} + : { task: false, task_status: false }), ...Object.fromEntries((cfg.experimental?.primary_tools ?? []).map((item) => [item, false])), }, - parts: promptParts, }) - if (resolvedOutputSchema) { - const structured = result.info.role === "assistant" ? result.info.structured : undefined - if (structured !== undefined) return JSON.stringify(structured) - } - return result.parts.findLast((item) => item.type === "text")?.text ?? "" }) const runTask = Effect.fn("TaskTool.runTask")(function* () { - return yield* TaskConcurrency.withTaskSlot({ - parentSessionID: ctx.sessionID, - subagentType: params.subagent_type, - agentMaxConcurrency, - caps, - effect: runTaskInner(), - }) + return yield* withTaskRunLease( + activeRunState, + activeOwner, + TaskConcurrency.withTaskSlot({ + parentSessionID: ctx.sessionID, + subagentType: params.subagent_type, + agentMaxConcurrency, + caps, + effect: runTaskInner(), + }), + ).pipe(Effect.provideService(Database.Service, database)) }) const markFinished = Effect.fn("TaskTool.markSubagentFinished")(function* ( state: "completed" | "error" | "cancelled" | "interrupted", - reason?: "human" | "parent_interrupted" | "timeout" | "takeover" | "runtime_error", + reason?: SubagentTerminalReason, + details?: SettlementDetails, ) { - const current = yield* sessions.get(a.nextSession.id).pipe(Effect.orDie) - yield* sessions - .setMetadata({ - sessionID: a.nextSession.id, - metadata: { - ...(current.metadata ?? {}), - deepagent: { - ...((current.metadata?.["deepagent"] as Record | undefined) ?? {}), - // §4.3 SubagentRunMetadata: state is the durable authority for terminal state. - // Compat: old data using `finished: true` with state is still readable unchanged. - subagent: { finished: true, state, at: Date.now(), ...(reason ? { reason } : {}) }, - }, - }, - }) - .pipe(Effect.ignore) + const won = yield* settleSubagentRun( + sessions, + activeRunState, + activeOwner, + state, + reason ?? "runtime_error", + { + output: details?.output, + error: details?.error, + notification: details?.notifyText ? notification(details.notifyText) : undefined, + }, + ).pipe(Effect.provideService(Database.Service, database)) + if (!won) yield* Effect.fail(lostTaskRunLease(activeRunState)) }) const inject = Effect.fn("TaskTool.injectBackgroundResult")(function* ( - state: "completed" | "error", - text: string, - doneTakeovers: number, + _state: "completed" | "error", + _text: string, + _doneTakeovers: number, ) { - const currentParent = yield* sessions.get(ctx.sessionID) - const suffix = - doneTakeovers > 0 ? ` (after ${doneTakeovers} takeover${doneTakeovers === 1 ? "" : "s"})` : "" - yield* ops - .prompt({ - sessionID: ctx.sessionID, - agent: currentParent.agent ?? ctx.agent, - variant, - parts: [ - { - type: "text", - synthetic: true, - text: renderOutput({ - sessionID: a.nextSession.id, - state, - summary: - state === "completed" - ? `Background task completed: ${params.description}${suffix}` - : `Background task failed: ${params.description}${suffix}`, - text, - maxChars: flags.subagentOutputMaxChars, - }), - }, - ], - }) - .pipe(Effect.ignore, Effect.forkIn(scope, { startImmediately: true })) + yield* dispatchNotifications() }) // 1d: worktree teardown hangs off the completion points (only in this flag-gated path; the @@ -518,9 +1217,10 @@ export const TaskTool = Tool.define( if (!a.worktreeInfo) return const worktreeOpt = yield* Effect.serviceOption(Worktree.Service) if (Option.isNone(worktreeOpt)) return - yield* (force - ? worktreeOpt.value.remove({ directory: a.worktreeInfo.directory }) - : worktreeOpt.value.safeRemove({ directory: a.worktreeInfo.directory }) + yield* ( + force + ? worktreeOpt.value.remove({ directory: a.worktreeInfo.directory }) + : worktreeOpt.value.safeRemove({ directory: a.worktreeInfo.directory }) ).pipe(Effect.ignore) }) @@ -568,7 +1268,7 @@ export const TaskTool = Tool.define( const diag = result.type === "conflict" ? `conflicts in ${result.paths.join(", ")}` - : result.diagnostic ?? result.type + : (result.diagnostic ?? result.type) return yield* Effect.fail( new Error( `Automatic worktree merge failed (${diag}). ` + @@ -597,13 +1297,16 @@ export const TaskTool = Tool.define( type: id, title: params.description, metadata, - onPromote: Effect.all([ - ctx.metadata({ - title: params.description, - metadata: { ...metadata, background: true, jobId: a.nextSession.id }, - }), - driveBackground(bundle, takeovers), - ]), + onPromote: Effect.all( + [ + ctx.metadata({ + title: params.description, + metadata: { ...metadata, background: true, jobId: a.nextSession.id }, + }), + driveBackground(bundle, takeovers), + ], + { discard: true }, + ), run: runTask().pipe(Effect.onInterrupt(() => ops.cancel(a.nextSession.id))), }) return { kind: "started" as const, bundle } @@ -663,15 +1366,22 @@ export const TaskTool = Tool.define( Effect.catchCause((cause) => Effect.gen(function* () { const diagnostic = Cause.squash(cause) - yield* b.markFinished("error") + yield* b.markFinished("error", "runtime_error", { + error: { code: "runtime_error", message: String(diagnostic) }, + }) yield* b.inject("error", `Worktree merge-back failed: ${String(diagnostic)}`, takeovers) yield* b.teardownWorktree(false) return false }), ), ) - if (!merged && b.automaticWriteIsolation) return yield* Effect.fail(new Error("Worktree merge-back failed")) - yield* b.markFinished("completed") + if (!merged && b.automaticWriteIsolation) + return yield* Effect.fail(new Error("Worktree merge-back failed")) + yield* b.markFinished( + "completed", + resolvedOutputSchema ? "structured_output_valid" : "text_output_valid", + { output: outcome.output }, + ) yield* b.teardownWorktree(merged) return { title: params.description, @@ -700,24 +1410,37 @@ export const TaskTool = Tool.define( } if (takeovers >= takeoverLimit) { yield* background.cancel(b.nextSession.id).pipe(Effect.ignore) - yield* b.markFinished("error") + const reason = outcome.reason.startsWith("timed out") ? "timeout" : terminalReason(outcome.reason) + yield* b.markFinished("error", reason, { error: { code: reason, message: outcome.reason } }) yield* b.teardownWorktree(false) - return { - title: params.description, - metadata: b.metadata, - output: renderOutput({ + return yield* Effect.fail( + taskError({ + code: reason, + message: `The subagent failed after ${takeovers} bounded takeover attempt(s). Last failure: ${outcome.reason}`, sessionID: b.nextSession.id, - state: "error", - summary: `Background task failed: ${params.description} (after ${takeovers} takeover${takeovers === 1 ? "" : "s"})`, - text: `The subagent was retried ${takeovers} time(s) after timeout/crash and still did not complete. Last failure: ${outcome.reason}. The half-finished attempt was cancelled and its worktree discarded.`, - maxChars: flags.subagentOutputMaxChars, + phase: outcome.reason.includes("Phase: finalize") ? "finalize" : "research", + attempts: takeovers + 1, }), - } + ) } yield* background.cancel(b.nextSession.id).pipe(Effect.ignore) - yield* b.markFinished("cancelled") + yield* b.markFinished("cancelled", "takeover") yield* b.teardownWorktree(true) - const next = yield* startAttempt(yield* spawnAttempt(false), takeovers + 1, false) + const childSessionID = SessionID.create() + const takeoverRun = yield* spawnTaskTakeover({ root: admission.run, childSessionID }).pipe( + Effect.provideService(Database.Service, database), + ) + const claimedTakeover = yield* claimTaskProvisioning({ run: takeoverRun, owner: executionOwner }).pipe( + Effect.provideService(Database.Service, database), + ) + if (!claimedTakeover) + return yield* Effect.die(new Error(`Takeover run ${takeoverRun.runID} lost its provisioning claim`)) + const next = yield* startAttempt( + yield* spawnAttempt(false, claimedTakeover), + claimedTakeover, + takeovers + 1, + false, + ) if (next.kind === "extended") return yield* Effect.die(new Error("unreachable: extend on a fresh takeover attempt")) return yield* driveForeground(next.bundle, takeovers + 1) @@ -732,31 +1455,65 @@ export const TaskTool = Tool.define( Effect.catchCause((cause) => Effect.gen(function* () { const diagnostic = Cause.squash(cause) - yield* b.markFinished("error") - yield* b.inject("error", `Worktree merge-back failed: ${String(diagnostic)}`, takeovers) + const text = `Worktree merge-back failed: ${String(diagnostic)}` + yield* b.markFinished("error", "runtime_error", { + error: { code: "runtime_error", message: text }, + notifyText: renderOutput({ + sessionID: b.nextSession.id, + state: "error", + summary: `Background task failed: ${params.description}`, + text, + maxChars: flags.subagentOutputMaxChars, + }), + }) + yield* b.inject("error", text, takeovers) yield* b.teardownWorktree(false) return false }), ), ) if (!merged && b.automaticWriteIsolation) return - yield* b.markFinished("completed") + const output = waited.info?.output ?? "" + yield* b.markFinished( + "completed", + resolvedOutputSchema ? "structured_output_valid" : "text_output_valid", + { + output, + notifyText: renderOutput({ + sessionID: b.nextSession.id, + state: "completed", + summary: `Background task completed: ${params.description}${takeovers === 0 ? "" : ` (after ${takeovers} takeover${takeovers === 1 ? "" : "s"})`}`, + text: output, + maxChars: flags.subagentOutputMaxChars, + }), + }, + ) yield* b.teardownWorktree(merged) - yield* b.inject("completed", waited.info?.output ?? "", takeovers) + yield* b.inject("completed", output, takeovers) return } if (!waited.timedOut && status === "cancelled") { - yield* b.markFinished("cancelled") + yield* b.markFinished("cancelled", "parent_interrupted") yield* b.teardownWorktree(false) return } if (takeovers >= takeoverLimit) { yield* background.cancel(b.nextSession.id).pipe(Effect.ignore) - yield* b.markFinished("error") + const reason = waited.timedOut ? `timed out after ${timeoutMs}ms` : (waited.info?.error ?? "Task failed") + yield* b.markFinished("error", waited.timedOut ? "timeout" : terminalReason(waited.info?.error), { + error: { + code: waited.timedOut ? "timeout" : terminalReason(waited.info?.error), + message: reason, + }, + notifyText: renderOutput({ + sessionID: b.nextSession.id, + state: "error", + summary: `Background task failed: ${params.description}`, + text: `The subagent was retried ${takeovers} time(s) after timeout/crash and still did not complete. Last failure: ${reason}. The half-finished attempt was cancelled and its worktree discarded.`, + maxChars: flags.subagentOutputMaxChars, + }), + }) yield* b.teardownWorktree(false) - const reason = waited.timedOut - ? `timed out after ${timeoutMs}ms` - : (waited.info?.error ?? "Task failed") yield* b.inject( "error", `The subagent was retried ${takeovers} time(s) after timeout/crash and still did not complete. Last failure: ${reason}. The half-finished attempt was cancelled and its worktree discarded.`, @@ -765,15 +1522,30 @@ export const TaskTool = Tool.define( return } yield* background.cancel(b.nextSession.id).pipe(Effect.ignore) - yield* b.markFinished("cancelled") + yield* b.markFinished("cancelled", "takeover") yield* b.teardownWorktree(true) - const next = yield* startAttempt(yield* spawnAttempt(false), takeovers + 1, false) + const childSessionID = SessionID.create() + const takeoverRun = yield* spawnTaskTakeover({ root: admission.run, childSessionID }).pipe( + Effect.provideService(Database.Service, database), + ) + const claimedTakeover = yield* claimTaskProvisioning({ run: takeoverRun, owner: executionOwner }).pipe( + Effect.provideService(Database.Service, database), + ) + if (!claimedTakeover) + return yield* Effect.die(new Error(`Takeover run ${takeoverRun.runID} lost its provisioning claim`)) + const next = yield* startAttempt( + yield* spawnAttempt(false, claimedTakeover), + claimedTakeover, + takeovers + 1, + false, + ) if (next.kind === "extended") return yield* Effect.die(new Error("unreachable: extend on a fresh takeover attempt")) yield* driveBackground(next.bundle, takeovers + 1) }).pipe(Effect.forkIn(scope, { startImmediately: true }), Effect.asVoid) - const started = yield* startAttempt(yield* spawnAttempt(true), 0, true) + const initialRun = claimedRun ?? admission.run + const started = yield* startAttempt(yield* spawnAttempt(true, initialRun), initialRun, 0, true) if (started.kind === "extended") { return { title: params.description, @@ -807,7 +1579,7 @@ export const TaskTool = Tool.define( // (unique even within the same millisecond) guarantees no two invocations request the same name. // Only the non-git degradation (NotGitError) is tolerated as a shared-directory fallback; any // other create failure now FAILS the task loudly instead of silently un-isolating it. - const isolate = !session && (params.isolation === "worktree" || subagentIsWriteType(next)) + const isolate = !admittedSession && (params.isolation === "worktree" || subagentIsWriteType(next)) const worktreeOpt = isolate ? yield* Effect.serviceOption(Worktree.Service) : Option.none() const worktreeInfo = isolate && Option.isSome(worktreeOpt) @@ -817,8 +1589,9 @@ export const TaskTool = Tool.define( : undefined const nextSession = - session ?? + admittedSession ?? (yield* sessions.create({ + id: admission.run.childSessionID, parentID: ctx.sessionID, title: params.description + ` (@${next.name} subagent)`, agent: next.name, @@ -845,17 +1618,6 @@ export const TaskTool = Tool.define( ], })) - const msg = yield* MessageV2.get({ sessionID: ctx.sessionID, messageID: ctx.messageID }).pipe( - Effect.provideService(Database.Service, database), - Effect.orDie, - ) - if (msg.info.role !== "assistant") return yield* Effect.fail(new Error("Not an assistant message")) - const variant = msg.info.variant - - const model = next.model ?? { - modelID: msg.info.modelID, - providerID: msg.info.providerID, - } const metadata = { parentSessionId: ctx.sessionID, sessionId: nextSession.id, @@ -866,154 +1628,71 @@ export const TaskTool = Tool.define( ...(runInBackground ? { background: true } : {}), } - // Subagent work-intensity: "downgrade" runs each child exactly one strength below the parent's - // EFFECTIVE agentMode (ultra→max→…→general; general stays general); "inherit" (default) leaves - // the child on the process-global mode. The chosen mode is injected ONLY as this child session's - // first user-message metadata (`deepagent.agent_mode_override`) — a per-request channel that - // request.ts reads per prompt and never touches the process-global agentMode, so concurrent - // subagents stay isolated from each other. "inherit" injects nothing (natural global inheritance). - const subagentIntensity = - (cfg.provider?.deepagent?.options?.subagentIntensity as string | undefined) === "downgrade" - ? "downgrade" - : "inherit" - const childAgentModeOverride = - subagentIntensity === "downgrade" ? downgradeOneLevel(AgentGateway.snapshot().agentMode) : undefined - yield* ctx.metadata({ title: params.description, metadata, }) - - const ops = ctx.extra?.promptOps as TaskPromptOps - if (!ops) return yield* Effect.fail(new Error("TaskTool requires promptOps in ctx.extra")) - - // L3 (路 B): when the caller supplied output_schema, drive the subagent's final turn through - // the structured-output path so the result parses deterministically. Undefined ⇒ free-text. - const resolvedOutputSchema = resolveOutputSchema(params.output_schema, params.subagent_type) - - // §5a: resolve the CODE-layer orchestration caps (configurable, lenient defaults). The - // per-parent-session semaphore below is the hard concurrency gate for parallel `task` calls; - // §C.3 agent `limits.maxConcurrency` tightens it further (min) when the subagent declares one. - const caps: CoreOrchestration.OrchestrationCaps = { - maxFanout: cfg.experimental?.orchestration?.max_fanout, - maxConcurrency: cfg.experimental?.orchestration?.max_concurrency, - } - const agentMaxConcurrency = next.limits?.maxConcurrency + const runState = yield* activateRun(claimedRun ?? admission.run) + const runOwner = runState.executionOwner ?? executionOwner const runTask = Effect.fn("TaskTool.runTask")(function* () { // §5a chokepoint: BOTH the foreground and background dispatch paths route the subagent's // actual work through `runTask`, so acquiring the concurrency slot HERE bounds how many // subagents of this parent session execute in parallel — regardless of how many the model // fanned out in one message. Ordinary tools never reach this code path. - return yield* TaskConcurrency.withTaskSlot({ - parentSessionID: ctx.sessionID, - subagentType: params.subagent_type, - agentMaxConcurrency, - caps, - effect: runTaskInner(), - }) + return yield* withTaskRunLease( + runState, + runOwner, + TaskConcurrency.withTaskSlot({ + parentSessionID: ctx.sessionID, + subagentType: params.subagent_type, + agentMaxConcurrency, + caps, + effect: runTaskInner(), + }), + ).pipe(Effect.provideService(Database.Service, database)) }) const runTaskInner = Effect.fn("TaskTool.runTaskInner")(function* () { - const parts = yield* ops.resolvePromptParts(params.prompt) - // U5: when isolated in a worktree, tell the subagent it inherited the parent's context but - // operates in a separate checkout — paths translate, and it must re-read files before editing. - const promptParts = worktreeInfo - ? [ - { - type: "text" as const, - text: - `You are running in an ISOLATED git worktree at ${worktreeInfo.directory} (branch ${worktreeInfo.branch ?? "detached"}). ` + - `You inherited context from the parent session, but your working directory is this worktree. ` + - `Re-read files before editing (do not trust remembered paths/contents), and know your changes stay isolated until merged back.`, - }, - ...parts, - ] - : parts - const result = yield* ops.prompt({ - messageID: MessageID.ascending(), + return yield* runSubagentPrompt({ + ops, + prompt: params.prompt, sessionID: nextSession.id, - model: { - modelID: model.modelID, - providerID: model.providerID, - }, + model, variant: next.model ? undefined : variant, agent: next.name, - ...(childAgentModeOverride - ? { metadata: { deepagent: { agent_mode_override: childAgentModeOverride } } } - : {}), - // Build a Format INSTANCE (not a plain literal): OutputFormatJsonSchema is a Schema.Class - // whose encoder is instanceof-gated, so a plain object would fail when the message Info is - // serialized onto the MessageUpdated sync event. prompt() also normalizes defensively. - ...(resolvedOutputSchema - ? { format: new SessionV1.OutputFormatJsonSchema({ type: "json_schema", schema: resolvedOutputSchema }) } - : {}), + agentModeOverride: childAgentModeOverride, + outputSchema: resolvedOutputSchema, + runID: runState.runID, + budget: researchBudget, + worktreeInfo, + onResearchCompleted: (messageID) => + markSubagentResearchCompleted(runState, runOwner, messageID).pipe( + Effect.provideService(Database.Service, database), + ), + onFinalizing: (progress) => + markSubagentFinalizing(sessions, runState, runOwner, progress).pipe( + Effect.provideService(Database.Service, database), + ), + onFinalized: (messageID) => + markSubagentFinalized(runState, runOwner, messageID).pipe( + Effect.provideService(Database.Service, database), + ), tools: { - // F5: use evaluatePermission (not presence check) — "ask" is not sufficient, only - // an explicit "allow" rule on the subagent's own permission makes these tools visible. ...(evaluatePermission("todowrite", "*", next.permission).action === "allow" ? {} : { todowrite: false }), - // task_status is gated on the same `task` permission (see the takeover-path block above). - ...(evaluatePermission(id, "*", next.permission).action === "allow" ? {} : { task: false, task_status: false }), + ...(evaluatePermission(id, "*", next.permission).action === "allow" + ? {} + : { task: false, task_status: false }), ...Object.fromEntries((cfg.experimental?.primary_tools ?? []).map((item) => [item, false])), }, - parts: promptParts, }) - // L3 (路 B): with output_schema, the structured object is surfaced on the assistant message's - // `structured` field (via the forced StructuredOutput tool). Return it as JSON so the parent - // agent parses a guaranteed-conformant result. Fall back to the free-text extraction only when - // no schema was requested — that path (the brittle `findLast(text)`) is UNCHANGED for callers - // that don't pass output_schema. - if (resolvedOutputSchema) { - const structured = result.info.role === "assistant" ? result.info.structured : undefined - if (structured !== undefined) return JSON.stringify(structured) - // B3 fix: when the retry cap fired, the assistant message carries a StructuredOutputError. - // Silently returning "" here would make the parent agent see an empty success result and - // lose all signal that the subagent failed to produce structured output. Surface the error - // explicitly so the task tool's error path propagates it correctly to the parent. - const msgError = result.info.role === "assistant" ? result.info.error : undefined - if (msgError && SessionV1.StructuredOutputError.isInstance(msgError)) { - return yield* Effect.fail( - new Error( - // U1: include the child session ID so the parent agent can recover partial work via - // task_read({ task_id: nextSession.id }) before retrying or duplicating the task. - `StructuredOutput failed (${msgError.data.retries} attempt(s)): ${msgError.data.message}. ` + - `Partial research is preserved in subagent session ${nextSession.id}. ` + - `Call task_read({ task_id: "${nextSession.id}" }) to recover completed work before retrying.`, - ), - ) - } - } - return result.parts.findLast((item) => item.type === "text")?.text ?? "" }) const inject = Effect.fn("TaskTool.injectBackgroundResult")(function* ( - state: "completed" | "error", - text: string, + _state: "completed" | "error", + _text: string, ) { - const currentParent = yield* sessions.get(ctx.sessionID) - yield* ops - .prompt({ - sessionID: ctx.sessionID, - agent: currentParent.agent ?? ctx.agent, - variant, - parts: [ - { - type: "text", - synthetic: true, - text: renderOutput({ - sessionID: nextSession.id, - state, - summary: - state === "completed" - ? `Background task completed: ${params.description}` - : `Background task failed: ${params.description}`, - text, - maxChars: flags.subagentOutputMaxChars, - }), - }, - ], - }) - .pipe(Effect.ignore, Effect.forkIn(scope, { startImmediately: true })) + yield* dispatchNotifications() }) // Mark the subagent session as FINISHED once it completes a run. A subagent does exactly one @@ -1027,34 +1706,46 @@ export const TaskTool = Tool.define( // Read-merge because setMetadata replaces the whole metadata object. const markFinished = Effect.fn("TaskTool.markSubagentFinished")(function* ( state: "completed" | "error" | "cancelled" | "interrupted", - reason?: "human" | "parent_interrupted" | "timeout" | "takeover" | "runtime_error", + reason?: SubagentTerminalReason, + details?: SettlementDetails, ) { - // Resume (`params.task_id`) reuses an existing session; a fresh finish marker is still correct - // (the reused session just completed another turn), so no special-casing is needed. - const current = yield* sessions.get(nextSession.id).pipe(Effect.orDie) - yield* sessions - .setMetadata({ - sessionID: nextSession.id, - metadata: { - ...(current.metadata ?? {}), - deepagent: { - ...((current.metadata?.["deepagent"] as Record | undefined) ?? {}), - // §4.3: state is the durable terminal-state authority; reason narrows the cause. - subagent: { finished: true, state, at: Date.now(), ...(reason ? { reason } : {}) }, - }, - }, - }) - .pipe(Effect.ignore) + const won = yield* settleSubagentRun(sessions, runState, runOwner, state, reason ?? "runtime_error", { + output: details?.output, + error: details?.error, + notification: details?.notifyText ? notification(details.notifyText) : undefined, + }).pipe(Effect.provideService(Database.Service, database)) + if (!won) yield* Effect.fail(lostTaskRunLease(runState)) }) const notify = Effect.fn("TaskTool.notifyBackgroundResult")(function* (jobID: string) { yield* background.wait({ id: jobID }).pipe( Effect.flatMap((result) => { if (result.info?.status === "completed") - return markFinished("completed").pipe(Effect.andThen(inject("completed", result.info.output ?? ""))) + return markFinished("completed", resolvedOutputSchema ? "structured_output_valid" : "text_output_valid", { + output: result.info.output ?? "", + notifyText: renderOutput({ + sessionID: nextSession.id, + state: "completed", + summary: `Background task completed: ${params.description}`, + text: result.info.output ?? "", + maxChars: flags.subagentOutputMaxChars, + }), + }).pipe(Effect.andThen(inject("completed", result.info.output ?? ""))) if (result.info?.status === "error") - return markFinished("error").pipe(Effect.andThen(inject("error", result.info.error ?? ""))) - if (result.info?.status === "cancelled") return markFinished("cancelled") + return markFinished("error", terminalReason(result.info.error), { + error: { + code: terminalReason(result.info.error), + message: result.info.error ?? "Task failed", + }, + notifyText: renderOutput({ + sessionID: nextSession.id, + state: "error", + summary: `Background task failed: ${params.description}`, + text: result.info.error ?? "Task failed", + maxChars: flags.subagentOutputMaxChars, + }), + }).pipe(Effect.andThen(inject("error", result.info.error ?? ""))) + if (result.info?.status === "cancelled") return markFinished("cancelled", "parent_interrupted") return Effect.void }), Effect.forkIn(scope, { startImmediately: true }), @@ -1084,13 +1775,16 @@ export const TaskTool = Tool.define( type: id, title: params.description, metadata, - onPromote: Effect.all([ - ctx.metadata({ - title: params.description, - metadata: { ...metadata, background: true, jobId: nextSession.id }, - }), - notify(nextSession.id), - ]), + onPromote: Effect.all( + [ + ctx.metadata({ + title: params.description, + metadata: { ...metadata, background: true, jobId: nextSession.id }, + }), + notify(nextSession.id), + ], + { discard: true }, + ), run: runTask().pipe(Effect.onInterrupt(() => ops.cancel(nextSession.id))), }) @@ -1137,7 +1831,9 @@ export const TaskTool = Tool.define( // Promoted to background: `notify` (registered via onPromote) owns the finish marker. if (result?.metadata?.background === true) return backgroundResult() if (result?.status === "error") { - yield* markFinished("error") + yield* markFinished("error", terminalReason(result.error), { + error: { code: terminalReason(result.error), message: result.error ?? "Task failed" }, + }) return yield* Effect.fail(new Error(result.error ?? "Task failed")) } if (result?.status === "cancelled") { @@ -1149,16 +1845,18 @@ export const TaskTool = Tool.define( ), ) } - yield* markFinished("completed") + yield* markFinished("completed", resolvedOutputSchema ? "structured_output_valid" : "text_output_valid", { + output: result?.output ?? "", + }) return { title: params.description, metadata, output: renderOutput({ - sessionID: nextSession.id, - state: "completed", - text: result?.output ?? "", - maxChars: flags.subagentOutputMaxChars, - }), + sessionID: nextSession.id, + state: "completed", + text: result?.output ?? "", + maxChars: flags.subagentOutputMaxChars, + }), } }), (_, exit) => diff --git a/packages/deepagent-code/src/tool/tool.ts b/packages/deepagent-code/src/tool/tool.ts index c8fcba7c..fae48fac 100644 --- a/packages/deepagent-code/src/tool/tool.ts +++ b/packages/deepagent-code/src/tool/tool.ts @@ -84,6 +84,8 @@ export interface Def< parameters: Parameters jsonSchema?: JSONSchema7 provenance?: Provenance + semanticFingerprint?(input: Schema.Schema.Type): unknown + resultFingerprint?(result: ExecuteResult): unknown execute(args: Schema.Schema.Type, ctx: Context): Effect.Effect> formatValidationError?(error: unknown): string } diff --git a/packages/deepagent-code/test/project/instance-bootstrap.test.ts b/packages/deepagent-code/test/project/instance-bootstrap.test.ts index 5ed9ba7d..4de3c663 100644 --- a/packages/deepagent-code/test/project/instance-bootstrap.test.ts +++ b/packages/deepagent-code/test/project/instance-bootstrap.test.ts @@ -10,6 +10,7 @@ import { InstanceStore } from "../../src/project/instance-store" import { disposeAllInstances, tmpdirScoped } from "../fixture/fixture" import { testEffect } from "../lib/effect" import { waitGlobalBusEvent } from "../server/global-bus" +import { registerDisposer, registerInitializer } from "@/effect/instance-registry" const it = testEffect(Layer.mergeAll(InstanceLayer.layer, CrossSpawnSpawner.defaultLayer)) @@ -73,14 +74,17 @@ it.live("InstanceStore.provide runs InstanceBootstrap before effect", () => }), ) -it.live("CLI bootstrap runs InstanceBootstrap before callback", () => - Effect.gen(function* () { - const tmp = yield* bootstrapFixture +it.live( + "CLI bootstrap runs InstanceBootstrap before callback", + () => + Effect.gen(function* () { + const tmp = yield* bootstrapFixture - yield* Effect.promise(() => cliBootstrap(tmp.directory, async () => "ok")) + yield* Effect.promise(() => cliBootstrap(tmp.directory, async () => "ok")) - expect(existsSync(tmp.marker)).toBe(true) - }), + expect(existsSync(tmp.marker)).toBe(true) + }), + 15_000, ) it.live("CLI bootstrap disposes the instance when the callback rejects", () => @@ -108,3 +112,41 @@ it.live("InstanceStore.reload runs InstanceBootstrap", () => expect(existsSync(tmp.marker)).toBe(true) }), ) + +it.live("InstanceStore runs registered recovery lifecycle with the resolved instance context", () => + Effect.gen(function* () { + const tmp = yield* bootstrapFixture + const store = yield* InstanceStore.Service + const initialized: string[] = [] + const disposed: string[] = [] + const unregisterInitializer = registerInitializer((ctx) => { + initialized.push(ctx.directory) + return Promise.resolve() + }) + const unregisterDisposer = registerDisposer((directory) => { + disposed.push(directory) + return Promise.resolve() + }) + + const ctx = yield* store.load({ directory: tmp.directory }) + yield* store.dispose(ctx) + unregisterInitializer() + unregisterDisposer() + + expect(initialized).toEqual([tmp.directory]) + expect(disposed).toEqual([tmp.directory]) + }), +) + +it.live("a failed recovery initializer is observable but does not brick workspace startup", () => + Effect.gen(function* () { + const tmp = yield* bootstrapFixture + const store = yield* InstanceStore.Service + const unregister = registerInitializer(() => Promise.reject(new Error("recovery unavailable"))) + + const ctx = yield* store.load({ directory: tmp.directory }) + unregister() + + expect(ctx.directory).toBe(tmp.directory) + }), +) diff --git a/packages/deepagent-code/test/session/message-v2.test.ts b/packages/deepagent-code/test/session/message-v2.test.ts index 9accf07b..395b4c19 100644 --- a/packages/deepagent-code/test/session/message-v2.test.ts +++ b/packages/deepagent-code/test/session/message-v2.test.ts @@ -684,6 +684,55 @@ describe("session.message-v2.toModelMessage", () => { ]) }) + test("preserves the custom tool marker across a model switch", async () => { + const assistantID = "m-assistant-custom" + const input: SessionV1.WithParts[] = [ + { + info: assistantInfo(assistantID, "m-user-custom", undefined, { providerID: "other", modelID: "other" }), + parts: [ + { + ...basePart(assistantID, "a-custom"), + type: "tool", + callID: "call-custom", + tool: "apply_patch", + state: { + status: "completed", + input: { patchText: "*** Begin Patch\n*** End Patch" }, + output: "Done", + title: "Apply patch", + metadata: {}, + time: { start: 0, end: 1 }, + }, + metadata: { deepagent: { toolType: "custom" } }, + }, + ] as SessionV1.Part[], + }, + ] + + expect(await MessageV2.toModelMessages(input, model)).toMatchObject([ + { + role: "assistant", + content: [ + { + type: "tool-call", + toolCallId: "call-custom", + providerOptions: { deepagent: { toolType: "custom" } }, + }, + ], + }, + { + role: "tool", + content: [ + { + type: "tool-result", + toolCallId: "call-custom", + providerOptions: { deepagent: { toolType: "custom" } }, + }, + ], + }, + ]) + }) + test("replaces compacted tool output with placeholder", async () => { const userID = "m-user" const assistantID = "m-assistant" diff --git a/packages/deepagent-code/test/session/processor-effect.test.ts b/packages/deepagent-code/test/session/processor-effect.test.ts index 68b0c907..831f759a 100644 --- a/packages/deepagent-code/test/session/processor-effect.test.ts +++ b/packages/deepagent-code/test/session/processor-effect.test.ts @@ -854,8 +854,9 @@ it.live("session.processor effect tests mark pending tools as aborted on cleanup expect(yield* llm.calls).toBe(1) expect(call?.state.status).toBe("error") if (call?.state.status === "error") { - expect(call.state.error).toBe("Tool execution aborted") + expect(call.state.error).toBe("Tool input was incomplete and was not executed") expect(call.state.metadata?.interrupted).toBe(true) + expect(call.state.metadata?.incompleteInput).toBe(true) expect(call.state.time.end).toBeDefined() } }), diff --git a/packages/deepagent-code/test/session/prompt.test.ts b/packages/deepagent-code/test/session/prompt.test.ts index f018f748..5259ca16 100644 --- a/packages/deepagent-code/test/session/prompt.test.ts +++ b/packages/deepagent-code/test/session/prompt.test.ts @@ -791,6 +791,46 @@ it.instance("loop continues (not exits) when finish is length: injects a continu }), ) +it.instance("loop retries truncated tool input with the bounded patch transaction guidance", () => + Effect.gen(function* () { + const { llm } = yield* useServerConfig(providerCfg) + const prompt = yield* SessionPrompt.Service + const sessions = yield* Session.Service + const session = yield* sessions.create({ + title: "Pinned", + permission: [{ permission: "*", pattern: "*", action: "allow" }], + }) + const seeded = yield* seed(session.id, { finish: "length" }) + yield* sessions.updatePart({ + id: PartID.ascending(), + messageID: seeded.assistant.id, + sessionID: session.id, + type: "tool", + callID: "truncated-call", + tool: "apply_patch", + state: { + status: "error", + input: {}, + error: "Tool input was incomplete and was not executed", + metadata: { interrupted: true, incompleteInput: true }, + time: { start: 1, end: 2 }, + }, + }) + yield* llm.text("completed with chunked patch") + + yield* prompt.loop({ sessionID: session.id }) + + const messages = yield* sessions.messages({ sessionID: session.id }) + expect( + messages.some((message) => + message.parts.some( + (part) => part.type === "text" && part.synthetic === true && part.text.includes("apply_patch_chunk"), + ), + ), + ).toBe(true) + }), +) + it.instance("glob tool keeps instance context during prompt runs", () => Effect.gen(function* () { const { dir, llm } = yield* useServerConfig(providerCfg) @@ -858,6 +898,69 @@ it.instance("loop continues when finish is stop but assistant has tool parts", ( }), ) +it.instance("non-interactive task token budget fails even when the provider stops naturally", () => + Effect.gen(function* () { + const { llm } = yield* useServerConfig(providerCfg) + const prompt = yield* SessionPrompt.Service + const sessions = yield* Session.Service + const session = yield* sessions.create({ title: "Pinned" }) + yield* prompt.prompt({ + sessionID: session.id, + agent: "build", + noReply: true, + metadata: { + deepagent: { + task_activity: { + interactive: false, + started_at: Date.now(), + budget: { max_steps: 4, max_tokens: 1, max_wall_ms: 60_000, max_no_progress: 2 }, + }, + }, + }, + parts: [{ type: "text", text: "bounded research" }], + }) + yield* llm.text("looks complete", { usage: { input: 2, output: 2 } }) + + const result = yield* prompt.loop({ sessionID: session.id }) + expect(result.info.role).toBe("assistant") + if (result.info.role === "assistant") expect(result.info.error?.name).toBe("TaskBudgetExceededError") + expect(yield* llm.calls).toBe(1) + }), +) + +it.instance("non-interactive task step budget prevents another provider turn", () => + Effect.gen(function* () { + const { llm } = yield* useServerConfig(providerCfg) + const prompt = yield* SessionPrompt.Service + const sessions = yield* Session.Service + const session = yield* sessions.create({ + title: "Pinned", + permission: [{ permission: "*", pattern: "*", action: "allow" }], + }) + yield* prompt.prompt({ + sessionID: session.id, + agent: "build", + noReply: true, + metadata: { + deepagent: { + task_activity: { + interactive: false, + started_at: Date.now(), + budget: { max_steps: 1, max_tokens: 10_000, max_wall_ms: 60_000, max_no_progress: 2 }, + }, + }, + }, + parts: [{ type: "text", text: "bounded research" }], + }) + yield* llm.tool("first", { value: "first" }) + + const result = yield* prompt.loop({ sessionID: session.id }) + expect(result.info.role).toBe("assistant") + if (result.info.role === "assistant") expect(result.info.error?.name).toBe("TaskBudgetExceededError") + expect(yield* llm.calls).toBe(1) + }), +) + it.instance("failed subtask preserves metadata on error tool state", () => Effect.gen(function* () { const { llm } = yield* useServerConfig((url) => ({ @@ -1999,6 +2102,62 @@ noLLMServer.instance( // Missing file handling +noLLMServer.instance( + "task notification prompt retries reuse the persisted user message", + () => + Effect.gen(function* () { + const prompt = yield* SessionPrompt.Service + const sessions = yield* Session.Service + const session = yield* sessions.create({}) + const messageID = MessageID.ascending() + const metadata = { + deepagent: { + task_notification: { run_id: "job_exact", outbox_id: "task-notify:job_exact" }, + }, + } + + const first = yield* prompt.prompt({ + sessionID: session.id, + messageID, + agent: "build", + noReply: true, + metadata, + parts: [{ type: "text", text: "task completed" }], + }) + const retry = yield* prompt.prompt({ + sessionID: session.id, + messageID, + agent: "build", + noReply: true, + metadata, + parts: [{ type: "text", text: "this payload must not be persisted twice" }], + }) + const conflict = yield* Effect.exit( + prompt.prompt({ + sessionID: session.id, + messageID, + agent: "build", + noReply: true, + metadata: { + deepagent: { + task_notification: { run_id: "job_other", outbox_id: "task-notify:job_other" }, + }, + }, + parts: [{ type: "text", text: "conflicting retry" }], + }), + ) + const stored = yield* MessageV2.get({ sessionID: session.id, messageID }) + + expect(first.info.id).toBe(messageID) + expect(retry).toEqual(first) + expect(stored.parts.filter((part) => part.type === "text").map((part) => part.text)).toEqual(["task completed"]) + expect(Exit.isFailure(conflict)).toBe(true) + + yield* sessions.remove(session.id) + }), + { config: cfg }, +) + noLLMServer.instance( "does not fail the prompt when a file part is missing", () => diff --git a/packages/deepagent-code/test/session/tool-sequence-tracker.test.ts b/packages/deepagent-code/test/session/tool-sequence-tracker.test.ts index eb3d1768..2d6360e0 100644 --- a/packages/deepagent-code/test/session/tool-sequence-tracker.test.ts +++ b/packages/deepagent-code/test/session/tool-sequence-tracker.test.ts @@ -346,6 +346,55 @@ describe("canonical JSON fingerprint (key order independence)", () => { }) }) +describe("tool-defined semantic fingerprints", () => { + test("display-only shell descriptions cannot evade period-1 detection", () => { + const t = new ToolSequenceTracker() + t.setFingerprintResolver((_tool, input) => { + const value = input as { command: string; workdir?: string; timeout?: number } + return { command: value.command, workdir: value.workdir ?? null, timeout: value.timeout ?? 120_000 } + }) + const inputs = ["prepare", "research complete", "submit next"].map((description) => ({ + command: "true", + description, + })) + t.push("1", t.fingerprint("bash", inputs[0])) + t.markDone("1") + t.push("2", t.fingerprint("bash", inputs[1])) + t.markDone("2") + t.push("3", t.fingerprint("bash", inputs[2])) + expect(t.detect()?.period).toBe(1) + }) + + test("equivalent result fingerprints ignore input changes and reset on observable progress", () => { + const t = new ToolSequenceTracker() + t.setResultFingerprintResolver((_tool, result) => result) + const unchanged = { snapshot: "tree-a", plan: { active: "step-1" } } + + t.push("1", t.fingerprint("bash", { command: "true" })) + expect(t.markDone("1", "bash", { exit: 0, output: "same" }, unchanged)?.count).toBe(1) + t.push("2", t.fingerprint("bash", { command: "printf ''" })) + expect(t.markDone("2", "bash", { exit: 0, output: "same" }, unchanged)?.count).toBe(2) + t.push("3", t.fingerprint("bash", { command: "touch artifact" })) + expect( + t.markDone("3", "bash", { exit: 0, output: "same" }, { snapshot: "tree-b", plan: unchanged.plan })?.count, + ).toBe(1) + t.push("4", t.fingerprint("bash", { command: "true" })) + expect( + t.markDone("4", "bash", { exit: 0, output: "same" }, { snapshot: "tree-b", plan: unchanged.plan })?.count, + ).toBe(2) + t.push("5", t.fingerprint("bash", { command: "true" })) + expect(t.markDone("5", "bash", { exit: 0, output: "changed" }, unchanged)?.count).toBe(1) + }) + + test("tools without a result hook do not claim no-progress evidence", () => { + const t = new ToolSequenceTracker() + t.push("1", t.fingerprint("read", { path: "/dynamic" })) + expect(t.markDone("1", "read", "first")).toBeUndefined() + t.push("2", t.fingerprint("read", { path: "/dynamic" })) + expect(t.markDone("2", "read", "first")).toBeUndefined() + }) +}) + // --------------------------------------------------------------------------- // "Prior calls must be done" invariant // --------------------------------------------------------------------------- diff --git a/packages/deepagent-code/test/tool/task-finalizer.test.ts b/packages/deepagent-code/test/tool/task-finalizer.test.ts new file mode 100644 index 00000000..569b6aa3 --- /dev/null +++ b/packages/deepagent-code/test/tool/task-finalizer.test.ts @@ -0,0 +1,231 @@ +import { describe, expect, test } from "bun:test" +import { SessionV1 } from "@deepagent-code/core/v1/session" +import { ModelV2 } from "@deepagent-code/core/model" +import { ProviderV2 } from "@deepagent-code/core/provider" +import { Effect } from "effect" +import type { SessionPrompt } from "@/session/prompt" +import { MessageID, PartID, SessionID } from "@/session/schema" +import { runSubagentPrompt, type SubagentPromptInput, type TaskPromptOps } from "@/tool/task" + +const model = { + modelID: ModelV2.ID.make("test-model"), + providerID: ProviderV2.ID.make("test-provider"), +} +const sessionID = SessionID.make("ses_task_finalizer") +const schema = { + type: "object", + properties: { result: { type: "string" } }, + required: ["result"], + additionalProperties: false, +} + +function response( + input: SessionPrompt.PromptInput, + options: { text?: string; structured?: unknown; error?: SessionV1.Assistant["error"] }, +): SessionV1.WithParts { + const id = MessageID.ascending() + return { + info: { + id, + parentID: input.messageID ?? MessageID.ascending(), + sessionID: input.sessionID, + role: "assistant", + mode: input.agent ?? "researcher", + agent: input.agent ?? "researcher", + path: { cwd: "/tmp", root: "/tmp" }, + cost: 0, + tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } }, + modelID: input.model?.modelID ?? model.modelID, + providerID: input.model?.providerID ?? model.providerID, + time: { created: Date.now() }, + finish: "stop", + ...(options.structured === undefined ? {} : { structured: options.structured }), + ...(options.error ? { error: options.error } : {}), + }, + parts: options.text + ? [ + { + id: PartID.ascending(), + messageID: id, + sessionID: input.sessionID, + type: "text", + text: options.text, + }, + ] + : [], + } +} + +function input(ops: TaskPromptOps, outputSchema: Record | undefined = schema): SubagentPromptInput { + return { + ops, + prompt: "research the subsystem", + sessionID, + model, + variant: "xhigh", + agent: "researcher", + agentModeOverride: undefined, + outputSchema, + tools: { task: false }, + worktreeInfo: undefined, + } +} + +function ops(prompt: TaskPromptOps["prompt"]): TaskPromptOps { + return { + cancel: () => Effect.void, + resolvePromptParts: (text) => Effect.succeed([{ type: "text" as const, text }]), + prompt, + } +} + +describe("task structured finalizer", () => { + test("schema-less tasks preserve the last-text compatibility path", async () => { + const calls: SessionPrompt.PromptInput[] = [] + const request = input( + ops((prompt) => + Effect.sync(() => { + calls.push(prompt) + const output = response(prompt, { text: "first" }) + output.parts.push({ + id: PartID.ascending(), + messageID: output.info.id, + sessionID: prompt.sessionID, + type: "text", + text: "last", + }) + return output + }), + ), + ) + request.outputSchema = undefined + const result = await Effect.runPromise(runSubagentPrompt(request)) + + expect(result).toBe("last") + expect(calls).toHaveLength(1) + }) + + test("research and finalization are separate durable prompts", async () => { + const calls: SessionPrompt.PromptInput[] = [] + const result = await Effect.runPromise( + runSubagentPrompt( + input( + ops((request) => + Effect.sync(() => { + calls.push(request) + return calls.length === 1 + ? response(request, { text: "persisted research" }) + : response(request, { structured: { result: "ok" } }) + }), + ), + ), + ), + ) + + expect(result).toBe('{"result":"ok"}') + expect(calls).toHaveLength(2) + expect(calls[0]?.format).toBeUndefined() + expect(calls[0]?.tools).toEqual({ task: false }) + expect(calls[0]?.metadata?.deepagent?.task_activity).toMatchObject({ + interactive: false, + budget: { max_steps: 64, max_tokens: 200_000, max_wall_ms: 1_800_000, max_no_progress: 6 }, + }) + expect(calls[1]?.format?.type).toBe("json_schema") + expect(calls[1]?.tools).toBeUndefined() + expect(calls[1]?.metadata?.deepagent).toMatchObject({ structured_finalizer: { attempt: 1 } }) + }) + + test("plain-text finalizer outcomes consume the two-attempt budget", async () => { + const calls: SessionPrompt.PromptInput[] = [] + const effect = runSubagentPrompt( + input( + ops((request) => + Effect.sync(() => { + calls.push(request) + return response(request, { text: calls.length === 1 ? "persisted research" : "plain text" }) + }), + ), + ), + ) + + await expect(Effect.runPromise(effect)).rejects.toThrow("[structured_output_missing]") + expect(calls).toHaveLength(3) + expect(calls.slice(1).map((call) => call.metadata?.deepagent?.structured_finalizer?.attempt)).toEqual([1, 2]) + }) + + test("assistant errors are propagated before any output fallback", async () => { + const calls: SessionPrompt.PromptInput[] = [] + const effect = runSubagentPrompt( + input( + ops((request) => + Effect.sync(() => { + calls.push(request) + return response(request, { + text: "must not become success", + error: new SessionV1.DoomLoopError({ + message: "repeated shell command", + tool: "bash", + period: 1, + count: 3, + }).toObject(), + }) + }), + ), + ), + ) + + await expect(Effect.runPromise(effect)).rejects.toThrow("[doom_loop]") + expect(calls).toHaveLength(1) + }) + + test("controller rejects structured values that fail the boundary schema", async () => { + const calls: SessionPrompt.PromptInput[] = [] + const effect = runSubagentPrompt( + input( + ops((request) => + Effect.sync(() => { + calls.push(request) + return calls.length === 1 + ? response(request, { text: "persisted research" }) + : response(request, { structured: { wrong: "field" } }) + }), + ), + ), + ) + + await expect(Effect.runPromise(effect)).rejects.toThrow("[structured_output_invalid]") + expect(calls).toHaveLength(3) + }) + + test("research wall-time exhaustion returns a recoverable typed task error", async () => { + const request = input(ops(() => Effect.never)) + request.budget = { maxSteps: 2, maxTokens: 100, maxWallMs: 5, maxNoProgress: 2 } + + await expect(Effect.runPromise(runSubagentPrompt(request))).rejects.toThrow("[budget_exhausted]") + await expect(Effect.runPromise(runSubagentPrompt(request))).rejects.toThrow( + `task_read({ task_id: "${sessionID}" })`, + ) + }) + + test("assistant budget errors are propagated before text fallback", async () => { + const effect = runSubagentPrompt( + input( + ops((request) => + Effect.succeed( + response(request, { + text: "must not become success", + error: new SessionV1.TaskBudgetExceededError({ + message: "token budget exhausted", + budget: "tokens", + limit: 10, + used: 11, + }).toObject(), + }), + ), + ), + ), + ) + + await expect(Effect.runPromise(effect)).rejects.toThrow("[budget_exhausted]") + }) +}) diff --git a/packages/deepagent-code/test/tool/task-run.test.ts b/packages/deepagent-code/test/tool/task-run.test.ts new file mode 100644 index 00000000..0cf81108 --- /dev/null +++ b/packages/deepagent-code/test/tool/task-run.test.ts @@ -0,0 +1,489 @@ +import { describe, expect } from "bun:test" +import { Database } from "@deepagent-code/core/database/database" +import { ProjectV2 } from "@deepagent-code/core/project" +import { ProjectTable } from "@deepagent-code/core/project/sql" +import { AbsolutePath } from "@deepagent-code/core/schema" +import { SessionTable, TaskNotificationOutboxTable, TaskRunTable } from "@deepagent-code/core/session/sql" +import { Effect, Layer } from "effect" +import { count, eq } from "drizzle-orm" +import { MessageID, SessionID } from "../../src/session/schema" +import { + acknowledgeTaskNotification, + admitTaskRun, + claimTaskNotifications, + claimTaskProvisioning, + getActiveTaskRunByChild, + markTaskFinalized, + markTaskFinalizing, + markTaskResearchCompleted, + rejectTaskNotification, + recoverExpiredTaskRuns, + renewTaskRunLease, + requestHash, + settleTaskRun, + startTaskRun, +} from "../../src/tool/task-run" +import { testEffect } from "../lib/effect" +import { tmpdirScoped } from "../fixture/fixture" +import { CrossSpawnSpawner } from "@deepagent-code/core/cross-spawn-spawner" + +const database = Layer.mergeAll(Database.layerFromPath(":memory:"), CrossSpawnSpawner.defaultLayer) +const it = testEffect(database) +const parentSessionID = SessionID.make("ses_task_run_parent") + +const setup = Effect.gen(function* () { + const { db } = yield* Database.Service + yield* db + .insert(ProjectTable) + .values({ id: ProjectV2.ID.global, worktree: AbsolutePath.make("/project"), sandboxes: [] }) + .onConflictDoNothing() + .run() + .pipe(Effect.orDie) + yield* db + .insert(SessionTable) + .values({ + id: parentSessionID, + project_id: ProjectV2.ID.global, + slug: "task-run-parent", + directory: "/project", + title: "parent", + version: "test", + }) + .onConflictDoNothing() + .run() + .pipe(Effect.orDie) +}) + +const admit = (input?: { + messageID?: MessageID + callID?: string + childSessionID?: SessionID + joinRunID?: string + request?: unknown + deliveryMode?: "foreground" | "background" + now?: number +}) => + admitTaskRun({ + parentSessionID, + parentMessageID: input?.messageID ?? MessageID.ascending("msg_task_run"), + toolCallID: input?.callID ?? "call-1", + childSessionID: input?.childSessionID, + joinRunID: input?.joinRunID, + request: input?.request ?? { prompt: "research", subagent_type: "researcher" }, + deliveryMode: input?.deliveryMode ?? "foreground", + now: input?.now, + }) + +describe("TaskRun durable store", () => { + it.effect("canonical request hashes ignore object key order", () => + Effect.sync(() => { + expect(requestHash({ b: [2, { y: true, x: null }], a: 1 })).toBe( + requestHash({ a: 1, b: [2, { x: null, y: true }] }), + ) + expect(requestHash({ prompt: "a" })).not.toBe(requestHash({ prompt: "b" })) + }), + ) + + it.effect("exact retry returns the original run and rejects conflicting reuse", () => + Effect.gen(function* () { + yield* setup + const messageID = MessageID.ascending("msg_exact_retry") + const first = yield* admit({ messageID, callID: "call-exact" }) + const retry = yield* admit({ messageID, callID: "call-exact" }) + + expect(first.exactRetry).toBe(false) + expect(first.runCreated).toBe(true) + expect(retry.exactRetry).toBe(true) + expect(retry.run.runID).toBe(first.run.runID) + expect(retry.run.childSessionID).toBe(first.run.childSessionID) + + const requestConflict = yield* Effect.flip( + admit({ messageID, callID: "call-exact", request: { prompt: "different" } }), + ) + expect(requestConflict.reason).toBe("request") + + const deliveryConflict = yield* Effect.flip( + admit({ messageID, callID: "call-exact", deliveryMode: "background" }), + ) + expect(deliveryConflict.reason).toBe("delivery") + }), + ) + + it.effect("concurrent database connections admit and settle exactly once", () => + Effect.gen(function* () { + const directory = yield* tmpdirScoped() + const filename = `${directory}/task-run.sqlite` + const databases = yield* Effect.all( + [Database.layerFromPath(filename), Database.layerFromPath(filename)].map((layer) => + Layer.build(layer.pipe(Layer.provide(CrossSpawnSpawner.defaultLayer))), + ), + { concurrency: "unbounded" }, + ) + yield* setup.pipe(Effect.provide(databases[0])) + const messageID = MessageID.ascending("msg_concurrent_admission") + const admissions = yield* Effect.all( + Array.from({ length: 12 }, (_, index) => + admit({ messageID, callID: "call-concurrent" }).pipe(Effect.provide(databases[index % databases.length])), + ), + { concurrency: "unbounded" }, + ) + const run = admissions[0].run + expect(new Set(admissions.map((item) => item.run.runID)).size).toBe(1) + expect(admissions.filter((item) => item.runCreated)).toHaveLength(1) + + const claimed = yield* claimTaskProvisioning({ run, owner: "worker", now: 1 }).pipe(Effect.provide(databases[0])) + const running = yield* startTaskRun(claimed!, "worker", 2).pipe(Effect.provide(databases[0])) + const settlements = yield* Effect.all( + [ + settleTaskRun({ + run: running!, + owner: "worker", + state: "completed", + reason: "structured_output_valid", + output: '{"result":"ok"}', + notification: { directory: "/project", payload: { agent: "build", text: "complete" } }, + now: 3, + }).pipe(Effect.provide(databases[0])), + settleTaskRun({ + run: running!, + owner: "worker", + state: "error", + reason: "provider_error", + error: { code: "provider_error", message: "late failure" }, + notification: { directory: "/project", payload: { agent: "build", text: "failed" } }, + now: 3, + }).pipe(Effect.provide(databases[1])), + ], + { concurrency: "unbounded" }, + ) + expect(settlements.filter((item) => item.won)).toHaveLength(1) + + const counts = yield* Effect.gen(function* () { + const { db } = yield* Database.Service + return { + runs: yield* db.select({ count: count() }).from(TaskRunTable).get().pipe(Effect.orDie), + notifications: yield* db + .select({ count: count() }) + .from(TaskNotificationOutboxTable) + .get() + .pipe(Effect.orDie), + } + }).pipe(Effect.provide(databases[0])) + expect(counts.runs).toEqual({ count: 1 }) + expect(counts.notifications).toEqual({ count: 1 }) + }), + ) + + it.effect("one active run per child, joined admissions share it, and later runs increment generation", () => + Effect.gen(function* () { + yield* setup + const childSessionID = SessionID.make("ses_task_run_child") + const first = yield* admit({ childSessionID, messageID: MessageID.ascending("msg_generation_1") }) + const claimed = yield* claimTaskProvisioning({ run: first.run, owner: "worker-1", now: 100, leaseMs: 50 }) + expect(claimed?.state).toBe("provisioning") + const running = yield* startTaskRun(claimed!, "worker-1", 101) + expect(running?.state).toBe("researching") + expect((yield* getActiveTaskRunByChild(childSessionID))?.runID).toBe(first.run.runID) + + const unjoined = yield* Effect.flip( + admit({ childSessionID, messageID: MessageID.ascending("msg_generation_conflict"), callID: "call-conflict" }), + ) + expect(unjoined.reason).toBe("join") + + const joined = yield* admit({ + childSessionID, + joinRunID: first.run.runID, + messageID: MessageID.ascending("msg_generation_join"), + callID: "call-join", + request: { prompt: "additional context" }, + }) + expect(joined.runCreated).toBe(false) + expect(joined.run.runID).toBe(first.run.runID) + + expect( + (yield* settleTaskRun({ + run: running!, + owner: "worker-1", + state: "completed", + reason: "text_output_valid", + output: "done", + now: 102, + })).won, + ).toBe(true) + expect(yield* getActiveTaskRunByChild(childSessionID)).toBeUndefined() + + const second = yield* admit({ + childSessionID, + messageID: MessageID.ascending("msg_generation_2"), + callID: "call-2", + }) + expect(second.run.generation).toBe(first.run.generation + 1) + expect(second.run.runID).not.toBe(first.run.runID) + }), + ) + + it.effect("provisioning lease permits only its owner until expiry", () => + Effect.gen(function* () { + yield* setup + const admission = yield* admit({ messageID: MessageID.ascending("msg_lease") }) + const first = yield* claimTaskProvisioning({ run: admission.run, owner: "worker-a", now: 1_000, leaseMs: 100 }) + expect(first?.executionOwner).toBe("worker-a") + expect( + yield* claimTaskProvisioning({ run: admission.run, owner: "worker-b", now: 1_099, leaseMs: 100 }), + ).toBeUndefined() + expect( + (yield* claimTaskProvisioning({ run: admission.run, owner: "worker-b", now: 1_100, leaseMs: 100 })) + ?.executionOwner, + ).toBe("worker-b") + expect(yield* startTaskRun(first!, "worker-a", 1_101)).toBeUndefined() + expect((yield* startTaskRun(admission.run, "worker-b", 1_102))?.state).toBe("researching") + }), + ) + + it.effect("expired provisioning settles before a late worker can create the child", () => + Effect.gen(function* () { + yield* setup + const admission = yield* admit({ + messageID: MessageID.ascending("msg_provisioning_recovery"), + deliveryMode: "background", + }) + const claimed = yield* claimTaskProvisioning({ run: admission.run, owner: "worker", now: 100, leaseMs: 50 }) + + expect(yield* recoverExpiredTaskRuns({ directory: "/project", now: 149 })).toEqual([]) + const recovered = yield* recoverExpiredTaskRuns({ directory: "/project", now: 150 }) + expect(recovered).toHaveLength(1) + expect(recovered[0].state).toBe("error") + expect(recovered[0].reason).toBe("execution_lease_expired") + expect(yield* startTaskRun(claimed!, "worker", 151)).toBeUndefined() + + const outbox = yield* Effect.gen(function* () { + const { db } = yield* Database.Service + return yield* db.select().from(TaskNotificationOutboxTable).get().pipe(Effect.orDie) + }) + expect(outbox?.payload.text).toContain("durable child session was being provisioned") + expect(outbox?.payload.text).not.toContain("Partial work is preserved") + }), + ) + + it.effect("running leases renew and expired runs settle once with an atomic recovery notification", () => + Effect.gen(function* () { + const directory = yield* tmpdirScoped() + const filename = `${directory}/task-run-recovery.sqlite` + const databases = yield* Effect.all( + [Database.layerFromPath(filename), Database.layerFromPath(filename)].map((layer) => + Layer.build(layer.pipe(Layer.provide(CrossSpawnSpawner.defaultLayer))), + ), + { concurrency: "unbounded" }, + ) + yield* setup.pipe(Effect.provide(databases[0])) + const admission = yield* admit({ messageID: MessageID.ascending("msg_recovery") }).pipe( + Effect.provide(databases[0]), + ) + const claimed = yield* claimTaskProvisioning({ run: admission.run, owner: "worker", now: 100, leaseMs: 50 }).pipe( + Effect.provide(databases[0]), + ) + const running = (yield* startTaskRun(claimed!, "worker", 101, 50).pipe(Effect.provide(databases[0])))! + + expect( + yield* renewTaskRunLease({ run: running, owner: "other", now: 140, leaseMs: 100 }).pipe( + Effect.provide(databases[1]), + ), + ).toBe(false) + expect( + yield* renewTaskRunLease({ run: running, owner: "worker", now: 140, leaseMs: 100 }).pipe( + Effect.provide(databases[0]), + ), + ).toBe(true) + expect( + yield* recoverExpiredTaskRuns({ directory: "/project", now: 239 }).pipe(Effect.provide(databases[1])), + ).toEqual([]) + expect( + yield* renewTaskRunLease({ run: running, owner: "worker", now: 240, leaseMs: 100 }).pipe( + Effect.provide(databases[0]), + ), + ).toBe(false) + expect( + (yield* settleTaskRun({ + run: running, + owner: "worker", + state: "completed", + reason: "text_output_valid", + output: "late result before recovery", + now: 240, + }).pipe(Effect.provide(databases[0]))).won, + ).toBe(false) + + const recovered = yield* Effect.all( + databases.map((database) => + recoverExpiredTaskRuns({ directory: "/project", now: 240 }).pipe(Effect.provide(database)), + ), + { concurrency: "unbounded" }, + ) + expect(recovered.flat()).toHaveLength(1) + expect(recovered.flat()[0].state).toBe("error") + expect(recovered.flat()[0].reason).toBe("execution_lease_expired") + expect( + (yield* settleTaskRun({ + run: running, + owner: "worker", + state: "completed", + reason: "text_output_valid", + output: "late result", + now: 241, + }).pipe(Effect.provide(databases[0]))).won, + ).toBe(false) + + const counts = yield* Effect.gen(function* () { + const { db } = yield* Database.Service + return { + runs: yield* db + .select({ count: count() }) + .from(TaskRunTable) + .where(eq(TaskRunTable.reason, "execution_lease_expired")) + .get() + .pipe(Effect.orDie), + notifications: yield* db + .select({ count: count() }) + .from(TaskNotificationOutboxTable) + .get() + .pipe(Effect.orDie), + } + }).pipe(Effect.provide(databases[1])) + expect(counts).toEqual({ runs: { count: 1 }, notifications: { count: 1 } }) + }), + ) + + it.effect("phase refs persist and settlement CAS has one immutable winner", () => + Effect.gen(function* () { + yield* setup + const admission = yield* admit({ messageID: MessageID.ascending("msg_settlement") }) + const claimed = yield* claimTaskProvisioning({ run: admission.run, owner: "worker", now: 10 }) + const running = (yield* startTaskRun(claimed!, "worker", 11))! + const rawMessageID = MessageID.ascending("msg_raw_result") + const finalMessageID = MessageID.ascending("msg_structured_result") + yield* markTaskResearchCompleted(running, "worker", rawMessageID, 12) + yield* markTaskFinalizing(running, "worker", 2, rawMessageID, 13) + yield* markTaskFinalized(running, "worker", finalMessageID, 14) + + const winner = yield* settleTaskRun({ + run: running, + owner: "worker", + state: "completed", + reason: "structured_output_valid", + output: '{"answer":"ok"}', + structuredResultMessageID: finalMessageID, + notification: { + directory: "/project", + payload: { agent: "build", text: "complete" }, + }, + now: 15, + }) + const loser = yield* settleTaskRun({ + run: running, + owner: "worker", + state: "error", + reason: "provider_error", + error: { code: "provider_error", message: "late failure" }, + now: 16, + }) + + expect(winner.won).toBe(true) + expect(winner.run.rawResultMessageID).toBe(rawMessageID) + expect(winner.run.structuredResultMessageID).toBe(finalMessageID) + expect(loser.won).toBe(false) + expect(loser.run.state).toBe("completed") + expect(loser.run.output).toBe('{"answer":"ok"}') + + const { db } = yield* Database.Service + expect(yield* db.select({ count: count() }).from(TaskNotificationOutboxTable).get().pipe(Effect.orDie)).toEqual({ + count: 1, + }) + }), + ) + + it.effect("outbox claims are leased, retry with backoff, and stale acknowledgements fail closed", () => + Effect.gen(function* () { + yield* setup + const admission = yield* admit({ messageID: MessageID.ascending("msg_outbox"), deliveryMode: "background" }) + const claimed = yield* claimTaskProvisioning({ run: admission.run, owner: "worker", now: 1 }) + const running = (yield* startTaskRun(claimed!, "worker", 2))! + yield* settleTaskRun({ + run: running, + owner: "worker", + state: "error", + reason: "provider_error", + error: { code: "provider_error", message: "provider unavailable" }, + notification: { + directory: "/project", + payload: { agent: "build", variant: "high", text: "failed" }, + }, + now: 10, + }) + + const first = (yield* claimTaskNotifications({ + owner: "dispatcher-a", + directory: "/project", + now: 10, + leaseMs: 50, + }))[0] + expect(first.attempts).toBe(1) + expect( + yield* claimTaskNotifications({ owner: "dispatcher-b", directory: "/project", now: 59, leaseMs: 50 }), + ).toEqual([]) + const reclaimed = (yield* claimTaskNotifications({ + owner: "dispatcher-a", + directory: "/project", + now: 60, + leaseMs: 50, + }))[0] + expect(reclaimed.attempts).toBe(2) + expect( + yield* acknowledgeTaskNotification({ + id: first.id, + owner: "dispatcher-a", + attempts: first.attempts, + now: 61, + }), + ).toBe(false) + expect( + yield* rejectTaskNotification({ + id: reclaimed.id, + owner: "dispatcher-a", + attempts: reclaimed.attempts, + error: "temporary", + now: 62, + }), + ).toBe(false) + expect(yield* claimTaskNotifications({ owner: "dispatcher-c", directory: "/project", now: 2_061 })).toEqual([]) + const third = (yield* claimTaskNotifications({ owner: "dispatcher-c", directory: "/project", now: 2_062 }))[0] + expect(third.attempts).toBe(3) + expect( + yield* acknowledgeTaskNotification({ + id: third.id, + owner: "dispatcher-c", + attempts: third.attempts, + now: 2_063, + }), + ).toBe(true) + expect(yield* claimTaskNotifications({ owner: "dispatcher-d", directory: "/project", now: 10_000 })).toEqual([]) + + const { db } = yield* Database.Service + expect( + (yield* db + .select({ status: TaskNotificationOutboxTable.status }) + .from(TaskNotificationOutboxTable) + .where(eq(TaskNotificationOutboxTable.id, third.id)) + .get() + .pipe(Effect.orDie))?.status, + ).toBe("delivered") + expect( + (yield* db + .select({ state: TaskRunTable.state }) + .from(TaskRunTable) + .where(eq(TaskRunTable.run_id, admission.run.runID)) + .get() + .pipe(Effect.orDie))?.state, + ).toBe("error") + }), + ) +}) diff --git a/packages/deepagent-code/test/tool/task-takeover.test.ts b/packages/deepagent-code/test/tool/task-takeover.test.ts index 2f8f2f06..c483cf0b 100644 --- a/packages/deepagent-code/test/tool/task-takeover.test.ts +++ b/packages/deepagent-code/test/tool/task-takeover.test.ts @@ -1,7 +1,7 @@ import { afterEach, describe, expect } from "bun:test" import { SessionV1 } from "@deepagent-code/core/v1/session" import { Database } from "@deepagent-code/core/database/database" -import { Effect, Exit, Layer } from "effect" +import { Cause, Effect, Exit, Layer } from "effect" import { mkdir } from "node:fs/promises" import { Agent } from "../../src/agent/agent" import { BackgroundJob } from "@/background/job" @@ -82,10 +82,7 @@ const takeoverBackgroundWorktree = testEffect( Layer.mergeAll(layer({ subagentTimeoutMs: 50, subagentTakeoverLimit: 2 }), worktreeMock), ) const e2e = testEffect( - Layer.mergeAll( - layer({ subagentTimeoutMs: 50, subagentTakeoverLimit: 2, subagentOutputMaxChars: 10 }), - worktreeMock, - ), + Layer.mergeAll(layer({ subagentTimeoutMs: 50, subagentTakeoverLimit: 2, subagentOutputMaxChars: 10 }), worktreeMock), ) const bounded = testEffect(layer({ subagentOutputMaxChars: 10 })) const off = testEffect(layer()) @@ -202,28 +199,31 @@ describe("tool.task takeover (v4.0.4 block1 1a+1b)", () => { }), ) - takeover.instance("a crashing subagent is retried and the retry result is delivered", () => - Effect.gen(function* () { - const { chat, assistant } = yield* seed() - const tool = yield* TaskTool - const def = yield* tool.init() - const calls: SessionID[] = [] - const promptOps = stubOps((input) => { - calls.push(input.sessionID) - if (calls.length === 1) return Effect.fail(new Error("boom")) - return Effect.succeed(reply(input, "ok after retry")) - }) - - const result = yield* def.execute( - { description: "inspect bug", prompt: "look into the cache key path", subagent_type: "general" }, - execCtx(chat, assistant, promptOps), - ) + takeover.instance( + "a crashing subagent is retried and the retry result is delivered", + () => + Effect.gen(function* () { + const { chat, assistant } = yield* seed() + const tool = yield* TaskTool + const def = yield* tool.init() + const calls: SessionID[] = [] + const promptOps = stubOps((input) => { + calls.push(input.sessionID) + if (calls.length === 1) return Effect.fail(new Error("boom")) + return Effect.succeed(reply(input, "ok after retry")) + }) + + const result = yield* def.execute( + { description: "inspect bug", prompt: "look into the cache key path", subagent_type: "general" }, + execCtx(chat, assistant, promptOps), + ) - expect(result.output).toContain(`state="completed"`) - expect(result.output).toContain("ok after retry") - expect(calls).toHaveLength(2) - expect(calls[0]).not.toBe(calls[1]) - }), + expect(result.output).toContain(`state="completed"`) + expect(result.output).toContain("ok after retry") + expect(calls).toHaveLength(2) + expect(calls[0]).not.toBe(calls[1]) + }), + 10_000, ) takeoverOnce.instance("exhausting the takeover limit surfaces a bounded failure to the parent", () => @@ -237,13 +237,18 @@ describe("tool.task takeover (v4.0.4 block1 1a+1b)", () => { return Effect.never }) - const result = yield* def.execute( - { description: "inspect bug", prompt: "look into the cache key path", subagent_type: "general" }, - execCtx(chat, assistant, promptOps), - ) + const exit = yield* def + .execute( + { description: "inspect bug", prompt: "look into the cache key path", subagent_type: "general" }, + execCtx(chat, assistant, promptOps), + ) + .pipe(Effect.exit) - expect(result.output).toContain(`state="error"`) - expect(result.output).toContain("takeover") + expect(Exit.isFailure(exit)).toBe(true) + const failure = Exit.isFailure(exit) ? Cause.pretty(exit.cause) : "" + expect(failure).toContain("[timeout]") + expect(failure).toContain("bounded takeover") + expect(failure).toContain("task_read") expect(calls).toHaveLength(2) const jobs = yield* BackgroundJob.Service @@ -263,17 +268,20 @@ describe("tool.task takeover (v4.0.4 block1 1a+1b)", () => { const def = yield* tool.init() const promptOps = stubOps(() => Effect.never) - const result = yield* def.execute( - { - description: "inspect bug", - prompt: "look into the cache key path", - subagent_type: "general", - isolation: "worktree", - }, - execCtx(chat, assistant, promptOps), - ) + const exit = yield* def + .execute( + { + description: "inspect bug", + prompt: "look into the cache key path", + subagent_type: "general", + isolation: "worktree", + }, + execCtx(chat, assistant, promptOps), + ) + .pipe(Effect.exit) - expect(result.output).toContain(`state="error"`) + expect(Exit.isFailure(exit)).toBe(true) + expect(Exit.isFailure(exit) ? Cause.pretty(exit.cause) : "").toContain("[timeout]") // one worktree per attempt (same fork base, fresh name), the first is force-recycled on // takeover, the last is teardown-safed when the limit is reached. expect(wt.created).toHaveLength(2) diff --git a/packages/deepagent-code/test/tool/task.test.ts b/packages/deepagent-code/test/tool/task.test.ts index 4a15c10f..c9562879 100644 --- a/packages/deepagent-code/test/tool/task.test.ts +++ b/packages/deepagent-code/test/tool/task.test.ts @@ -129,6 +129,7 @@ function reply(input: SessionPrompt.PromptInput, text: string): SessionV1.WithPa providerID: input.model?.providerID ?? ref.providerID, time: { created: Date.now() }, finish: "stop", + ...(input.format?.type === "json_schema" ? { structured: sampleSchema(input.format.schema) } : {}), }, parts: [ { @@ -142,6 +143,21 @@ function reply(input: SessionPrompt.PromptInput, text: string): SessionV1.WithPa } } +function sampleSchema(schema: Record): unknown { + if (Array.isArray(schema.enum) && schema.enum.length > 0) return schema.enum[0] + if (Array.isArray(schema.anyOf) && schema.anyOf.length > 0) + return sampleSchema(schema.anyOf[0] as Record) + if (schema.type === "object") { + const properties = schema.properties as Record> | undefined + return Object.fromEntries(Object.entries(properties ?? {}).map(([key, value]) => [key, sampleSchema(value)])) + } + if (schema.type === "array") return [] + if (schema.type === "number" || schema.type === "integer") return 0 + if (schema.type === "boolean") return true + if (schema.type === "null") return null + return "done" +} + describe("tool.task", () => { it.instance( "description sorts subagents by name and is stable across calls", @@ -537,8 +553,14 @@ describe("tool.task", () => { const { chat, assistant } = yield* seed() const tool = yield* TaskTool const def = yield* tool.init() - let seen: SessionPrompt.PromptInput | undefined - const promptOps = stubOps({ onPrompt: (input) => (seen = input) }) + let research: SessionPrompt.PromptInput | undefined + let finalizer: SessionPrompt.PromptInput | undefined + const promptOps = stubOps({ + onPrompt: (input) => { + if (input.format?.type === "json_schema") finalizer = input + else research = input + }, + }) const result = yield* def.execute( { @@ -561,6 +583,17 @@ describe("tool.task", () => { const child = yield* sessions.get(result.metadata.sessionId) expect(child.parentID).toBe(chat.id) expect(child.agent).toBe("reviewer") + expect(child.metadata?.deepagent?.subagent).toMatchObject({ + finished: true, + state: "completed", + phase: "settled", + reason: "structured_output_valid", + generation: 1, + attempts: 1, + run_id: expect.any(String), + raw_result_ref: expect.any(String), + settled_at: expect.any(Number), + }) expect(child.permission).toEqual([ { permission: "todowrite", @@ -578,11 +611,12 @@ describe("tool.task", () => { action: "allow", }, ]) - expect(seen?.tools).toEqual({ + expect(research?.tools).toEqual({ todowrite: false, bash: false, read: false, }) + expect(finalizer?.tools).toBeUndefined() }), { config: { @@ -739,6 +773,7 @@ describe("tool.task", () => { background.instance("background task completion waits for running updates", () => Effect.gen(function* () { const jobs = yield* BackgroundJob.Service + const sessions = yield* Session.Service const { chat, assistant } = yield* seed() const tool = yield* TaskTool const def = yield* tool.init() @@ -803,6 +838,11 @@ describe("tool.task", () => { const waited = yield* jobs.wait({ id: started.metadata.sessionId, timeout: 1_000 }) expect(waited.info?.status).toBe("completed") expect(waited.info?.output).toBe("second done") + expect((yield* sessions.get(started.metadata.sessionId)).metadata?.deepagent?.subagent).toMatchObject({ + generation: 1, + state: "completed", + finished: true, + }) const notification = yield* Effect.promise(() => injected.promise) expect(notification.variant).toBe("xhigh") expect(notification.parts[0]?.type).toBe("text") @@ -1081,7 +1121,11 @@ describe("tool.task", () => { AgentGateway.configure({ enabled: false, agentMode: "high", runsDir: undefined }) } }), - { config: { provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "downgrade" }, models: {} } } } }, + { + config: { + provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "downgrade" }, models: {} } }, + }, + }, ) // "inherit" (default): nothing is injected, so the child naturally runs at the process-global mode. @@ -1116,7 +1160,9 @@ describe("tool.task", () => { AgentGateway.configure({ enabled: false, agentMode: "high", runsDir: undefined }) } }), - { config: { provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "inherit" }, models: {} } } } }, + { + config: { provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "inherit" }, models: {} } } }, + }, ) // Per-request isolation: many concurrent downgraded subagents each carry their OWN override on their @@ -1169,6 +1215,10 @@ describe("tool.task", () => { AgentGateway.configure({ enabled: false, agentMode: "high", runsDir: undefined }) } }), - { config: { provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "downgrade" }, models: {} } } } }, + { + config: { + provider: { deepagent: { name: "DeepAgent", options: { subagentIntensity: "downgrade" }, models: {} } }, + }, + }, ) }) diff --git a/packages/desktop/package.json b/packages/desktop/package.json index d3299725..ed67945e 100644 --- a/packages/desktop/package.json +++ b/packages/desktop/package.json @@ -18,6 +18,8 @@ "preview": "electron-vite preview", "test": "bun test ./src", "test:ci": "mkdir -p .artifacts/unit && bun test ./src --reporter=junit --reporter-outfile=.artifacts/unit/junit.xml", + "test:subagents-smoke": "node --experimental-strip-types ./scripts/subagents-smoke.ts", + "test:subagents-sourcemap": "bun ./scripts/verify-subagents-sourcemap.ts", "package": "bun ./scripts/package.ts", "package:mac": "bun ./scripts/package.ts --mac", "package:win": "electron-builder --win --config electron-builder.config.ts", @@ -40,6 +42,8 @@ "@lydell/node-pty": "catalog:", "@deepagent-code/app": "workspace:*", "@deepagent-code/ui": "workspace:*", + "@jridgewell/trace-mapping": "0.3.31", + "@playwright/test": "catalog:", "@sentry/solid": "catalog:", "@sentry/vite-plugin": "catalog:", "@solid-primitives/i18n": "2.2.1", diff --git a/packages/desktop/scripts/subagents-smoke.ts b/packages/desktop/scripts/subagents-smoke.ts new file mode 100644 index 00000000..96b36544 --- /dev/null +++ b/packages/desktop/scripts/subagents-smoke.ts @@ -0,0 +1,189 @@ +#!/usr/bin/env bun +import { strict as assert } from "node:assert" +import { mkdir, mkdtemp, realpath, rm, stat } from "node:fs/promises" +import { tmpdir } from "node:os" +import { join, resolve } from "node:path" +import { _electron as electron, type ElectronApplication, type Page } from "@playwright/test" + +const root = await realpath(await mkdtemp(join(tmpdir(), "deepagent-code-subagents-smoke-"))) +const workspace = join(root, "workspace") +const main = resolve("out/main/index.js") +await mkdir(workspace, { recursive: true }) + +const env = Object.fromEntries( + Object.entries(process.env).filter((entry): entry is [string, string] => entry[1] !== undefined), +) +env.DEEPAGENT_CODE_TEST_ONBOARDING = "1" +env.DEEPAGENT_CODE_TEST_ROOT = root +env.DEEPAGENT_CODE_DB = join(root, "deepagent.sqlite") +env.DEEPAGENT_CODE_DISABLE_CHANNEL_DB = "1" + +type DesktopAPI = { + awaitInitialization(): Promise<{ url: string; username: string; password: string }> + storeGet(name: string, key: string): Promise + storeSet(name: string, key: string, value: string): Promise + getWindowCount(): Promise +} + +const api = (page: Page) => page.evaluate(() => (window as unknown as { api: DesktopAPI }).api.awaitInitialization()) + +let activeApp: ElectronApplication | undefined + +async function launch() { + const app = await electron.launch({ args: [main], env, timeout: 30_000 }) + activeApp = app + console.log( + "Electron main launched", + await app.evaluate(({ app, BrowserWindow }) => ({ + ready: app.isReady(), + windows: BrowserWindow.getAllWindows().length, + })), + ) + const page = await app.firstWindow({ timeout: 30_000 }) + await page.waitForFunction(() => Boolean((window as unknown as { api?: DesktopAPI }).api)) + return { app, page } +} + +async function close(app: ElectronApplication) { + await app.close() + if (activeApp === app) activeApp = undefined +} + +async function createSession( + server: Awaited>, + body: { title: string; parentID?: string; metadata?: Record }, +) { + const response = await fetch(`${server.url}/session?directory=${encodeURIComponent(workspace)}`, { + method: "POST", + headers: { + authorization: `Basic ${Buffer.from(`${server.username}:${server.password}`).toString("base64")}`, + "content-type": "application/json", + }, + body: JSON.stringify(body), + }) + if (!response.ok) throw new Error(`session create failed: ${response.status} ${await response.text()}`) + return (await response.json()) as { id: string } +} + +async function listSessions(server: Awaited>) { + const response = await fetch(`${server.url}/session?directory=${encodeURIComponent(workspace)}`, { + headers: { + authorization: `Basic ${Buffer.from(`${server.username}:${server.password}`).toString("base64")}`, + }, + }) + if (!response.ok) throw new Error(`session list failed: ${response.status} ${await response.text()}`) + return (await response.json()) as { id: string }[] +} + +try { + const first = await launch() + const server = await api(first.page) + const parent = await createSession(server, { title: "Cold-start parent" }) + const child = await createSession(server, { + title: "Cold-start researcher", + parentID: parent.id, + metadata: { + deepagent: { + subagent: { + finished: true, + state: "completed", + reason: "structured_output_valid", + run_id: "run_electron_smoke", + generation: 1, + }, + }, + }, + }) + const firstSessions = await listSessions(server) + assert.equal( + firstSessions.some((session) => session.id === parent.id), + true, + ) + assert.equal( + firstSessions.some((session) => session.id === child.id), + true, + ) + const slug = Buffer.from(workspace).toString("base64url") + const sessionKey = `local\u0000${slug}/${parent.id}` + await first.page.evaluate( + async ({ layout, pageLayout, server }) => { + const api = (window as unknown as { api: DesktopAPI }).api + await api.storeSet("deepagent.global.dat", "layout", JSON.stringify(layout)) + await api.storeSet("deepagent.global.dat", "layout.page", JSON.stringify(pageLayout)) + await api.storeSet("deepagent.global.dat", "server", JSON.stringify(server)) + }, + { + layout: { sessionView: { [sessionKey]: { scroll: {}, rightPanelMode: "subagents" } } }, + pageLayout: { + lastProjectSession: { + [workspace]: { directory: workspace, id: parent.id, at: Date.now() }, + }, + }, + server: { + list: [], + projects: { local: [{ worktree: workspace, expanded: true }] }, + lastProject: { local: workspace }, + }, + }, + ) + await close(first.app) + assert.equal((await stat(env.DEEPAGENT_CODE_DB)).size > 0, true, "desktop sidecar did not persist its database") + + const second = await launch() + const pageErrors: string[] = [] + const consoleErrors: string[] = [] + second.page.on("pageerror", (error) => pageErrors.push(error.stack ?? error.message)) + second.page.on("console", (message) => { + if (message.type() === "error") consoleErrors.push(message.text()) + }) + const sessions = await listSessions(await api(second.page)) + assert.equal( + sessions.some((session) => session.id === parent.id), + true, + "parent session did not survive restart", + ) + assert.equal( + sessions.some((session) => session.id === child.id), + true, + "child session did not survive restart", + ) + await second.page + .getByText("Cold-start researcher") + .waitFor({ state: "visible", timeout: 30_000 }) + .catch(async (error) => { + console.error("Electron cold-start diagnostics", { + body: (await second.page.locator("body").innerText()).slice(0, 4_000), + consoleErrors, + layout: await second.page.evaluate(() => + (window as unknown as { api: DesktopAPI }).api.storeGet("deepagent.global.dat", "layout"), + ), + pageLayout: await second.page.evaluate(() => + (window as unknown as { api: DesktopAPI }).api.storeGet("deepagent.global.dat", "layout.page"), + ), + server: await second.page.evaluate(() => + (window as unknown as { api: DesktopAPI }).api.storeGet("deepagent.global.dat", "server"), + ), + }) + throw error + }) + assert.equal(await second.page.getByRole("button", { name: /Cold-start researcher/ }).count(), 2) + assert.equal(await second.page.evaluate(() => (window as unknown as { api: DesktopAPI }).api.getWindowCount()), 1) + assert.deepEqual(pageErrors, []) + assert.deepEqual(consoleErrors, []) + assert.deepEqual( + await second.page.locator("#review-panel").evaluate((panel) => { + const selector = "a[href],button,input,select,textarea,summary,[role=button],[role=link],[role=tab]" + return [...panel.querySelectorAll(selector)] + .filter((element) => element.parentElement?.closest(selector)) + .map((element) => element.outerHTML) + }), + [], + ) + assert.equal(child.id.startsWith("ses_"), true) + await close(second.app) + console.log("Electron subagent cold-start smoke passed") +} finally { + if (activeApp) await activeApp.close().catch(() => undefined) + if (process.env.DEEPAGENT_CODE_KEEP_SMOKE === "1") console.error(`Electron smoke artifacts retained at ${root}`) + if (process.env.DEEPAGENT_CODE_KEEP_SMOKE !== "1") await rm(root, { recursive: true, force: true }) +} diff --git a/packages/desktop/scripts/verify-subagents-sourcemap.ts b/packages/desktop/scripts/verify-subagents-sourcemap.ts new file mode 100644 index 00000000..b91a373f --- /dev/null +++ b/packages/desktop/scripts/verify-subagents-sourcemap.ts @@ -0,0 +1,53 @@ +#!/usr/bin/env bun +import { strict as assert } from "node:assert" +import { basename, resolve } from "node:path" +import { generatedPositionFor, LEAST_UPPER_BOUND, originalPositionFor, TraceMap } from "@jridgewell/trace-mapping" + +const assets = resolve("out/renderer/assets") +const maps = await Array.fromAsync(new Bun.Glob("*.js.map").scan({ cwd: assets, absolute: true })) +const target = "side-panel-subagents.tsx" +const candidates = ( + await Promise.all( + maps.map(async (path) => { + const raw = (await Bun.file(path).json()) as { sources?: string[]; sourcesContent?: Array } + const index = raw.sources?.findIndex((source) => source.endsWith(target)) ?? -1 + return index === -1 ? undefined : { path, raw, index } + }), + ) +).filter((item) => item !== undefined) + +assert.equal(candidates.length, 1, `expected one production source map for ${target}, found ${candidates.length}`) +const candidate = candidates[0] +const source = candidate.raw.sources?.[candidate.index] +const content = candidate.raw.sourcesContent?.[candidate.index] +assert(source, `source entry missing from ${candidate.path}`) +assert(content, `sourcesContent entry missing from ${candidate.path}`) +assert.equal( + content, + await Bun.file(resolve("../app/src/pages/session/side-panel-subagents.tsx")).text(), + "production source map contains stale component source", +) + +const originalLine = content.split("\n").findIndex((line) => line.includes("export const SidePanelSubagents")) + 1 +assert(originalLine > 0, "SidePanelSubagents declaration missing from embedded source") + +const map = new TraceMap(await Bun.file(candidate.path).text()) +const generated = generatedPositionFor(map, { + source, + line: originalLine, + column: 0, + bias: LEAST_UPPER_BOUND, +}) +assert(generated.line !== null && generated.column !== null, "component declaration has no generated mapping") + +const stack = `at SidePanelSubagents (${basename(candidate.path, ".map")}:${generated.line}:${generated.column + 1})` +const frame = stack.match(/\((.+):(\d+):(\d+)\)$/) +assert(frame, `could not parse generated stack frame: ${stack}`) +const original = originalPositionFor(map, { + line: Number(frame[2]), + column: Number(frame[3]) - 1, +}) +assert(original.source?.endsWith(target), `stack mapped to unexpected source: ${original.source}`) +assert.equal(original.line, originalLine, `stack mapped to line ${original.line}, expected ${originalLine}`) + +console.log(`${stack} -> ${original.source}:${original.line}:${(original.column ?? 0) + 1}`) diff --git a/packages/desktop/src/main/index.ts b/packages/desktop/src/main/index.ts index ada580f4..777d3e41 100644 --- a/packages/desktop/src/main/index.ts +++ b/packages/desktop/src/main/index.ts @@ -118,12 +118,12 @@ const main = Effect.gen(function* () { const onboardingTestRoot = ((): string | undefined => { if (!TEST_ONBOARDING) return - const root = join(tmpdir(), `deepagent-code-onboarding-${randomUUID()}`) - rmSync(root, { recursive: true, force: true }) + const root = process.env.DEEPAGENT_CODE_TEST_ROOT ?? join(tmpdir(), `deepagent-code-onboarding-${randomUUID()}`) + if (!process.env.DEEPAGENT_CODE_TEST_ROOT) rmSync(root, { recursive: true, force: true }) ;["data", "config", "cache", "state", "desktop", "session"].forEach((dir) => mkdirSync(join(root, dir), { recursive: true }), ) - process.env.DEEPAGENT_CODE_DB = ":memory:" + process.env.DEEPAGENT_CODE_DB ??= ":memory:" process.env.XDG_DATA_HOME = join(root, "data") process.env.XDG_CONFIG_HOME = join(root, "config") process.env.XDG_CACHE_HOME = join(root, "cache") @@ -183,7 +183,9 @@ const main = Effect.gen(function* () { app.commandLine.appendSwitch("proxy-bypass-list", "<-loopback>") const features = app.commandLine.getSwitchValue("enable-features") app.commandLine.appendSwitch("enable-features", features ? `${jsCallStackFeature},${features}` : jsCallStackFeature) - if (!app.isPackaged) app.commandLine.appendSwitch("remote-debugging-port", "9222") + if (!app.isPackaged && !app.commandLine.hasSwitch("remote-debugging-port")) { + app.commandLine.appendSwitch("remote-debugging-port", "9222") + } if (!app.requestSingleInstanceLock()) { app.quit() From 2677717d29403b83c090b3ea0ccc5d727c7b9787 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:04:36 +0800 Subject: [PATCH 17/22] chore(sdk): regenerate clients and lockfile --- bun.lock | 7 ++- packages/sdk/js/src/gen/sdk.gen.ts | 38 ++++++++++++- packages/sdk/js/src/gen/types.gen.ts | 75 +++++++++++++++++++++++-- packages/sdk/js/src/v2/gen/sdk.gen.ts | 38 ++++++++++++- packages/sdk/js/src/v2/gen/types.gen.ts | 75 +++++++++++++++++++++++-- 5 files changed, 217 insertions(+), 16 deletions(-) diff --git a/bun.lock b/bun.lock index 78ae30c8..63533c10 100644 --- a/bun.lock +++ b/bun.lock @@ -29,7 +29,7 @@ }, "packages/app": { "name": "@deepagent-code/app", - "version": "1.4.0", + "version": "1.4.3", "dependencies": { "@codemirror/autocomplete": "6", "@codemirror/commands": "6", @@ -281,6 +281,7 @@ "@zip.js/zip.js": "2.7.62", "ai": "catalog:", "ai-gateway-provider": "3.1.2", + "ajv": "8.20.0", "bonjour-service": "1.3.0", "chokidar": "4.0.3", "cross-spawn": "catalog:", @@ -344,7 +345,7 @@ }, "packages/desktop": { "name": "@deepagent-code/desktop", - "version": "1.4.2", + "version": "1.4.3", "dependencies": { "@zip.js/zip.js": "2.7.62", "effect": "catalog:", @@ -359,7 +360,9 @@ "@actions/artifact": "4.0.0", "@deepagent-code/app": "workspace:*", "@deepagent-code/ui": "workspace:*", + "@jridgewell/trace-mapping": "0.3.31", "@lydell/node-pty": "catalog:", + "@playwright/test": "catalog:", "@sentry/solid": "catalog:", "@sentry/vite-plugin": "catalog:", "@solid-primitives/i18n": "2.2.1", diff --git a/packages/sdk/js/src/gen/sdk.gen.ts b/packages/sdk/js/src/gen/sdk.gen.ts index 17ff9ae9..dc7e7a68 100644 --- a/packages/sdk/js/src/gen/sdk.gen.ts +++ b/packages/sdk/js/src/gen/sdk.gen.ts @@ -88,6 +88,8 @@ import type { DeepagentKnowledgeRejectIdsErrors, DeepagentKnowledgeRejectIdsResponses, DeepagentKnowledgeRejectResponses, + DeepagentKnowledgeReviewSummaryErrors, + DeepagentKnowledgeReviewSummaryResponses, DeepagentKnowledgeShipGateErrors, DeepagentKnowledgeShipGateResponses, DeepagentPacksActiveErrors, @@ -2347,6 +2349,40 @@ export class Knowledge extends HeyApiClient { }) } + /** + * Get the DeepAgent knowledge review summary + * + * Return the pending review count without projecting the complete knowledge review list. + */ + public reviewSummary( + parameters?: { + directory?: string + workspace?: string + }, + options?: Options, + ) { + const params = buildClientParams( + [parameters], + [ + { + args: [ + { in: "query", key: "directory" }, + { in: "query", key: "workspace" }, + ], + }, + ], + ) + return (options?.client ?? this.client).get< + DeepagentKnowledgeReviewSummaryResponses, + DeepagentKnowledgeReviewSummaryErrors, + ThrowOnError + >({ + url: "/deepagent/knowledge/review-summary", + ...options, + ...params, + }) + } + /** * Approve DeepAgent knowledge by id * @@ -9059,7 +9095,7 @@ export class Session3 extends HeyApiClient { /** * Send message * - * Durably admit one session input and schedule agent-loop execution unless resume is false. + * Durably admit one session input when resume is false. This endpoint does not execute the production SessionPrompt engine. */ public prompt( parameters: { diff --git a/packages/sdk/js/src/gen/types.gen.ts b/packages/sdk/js/src/gen/types.gen.ts index ca9bc37f..44cf0622 100644 --- a/packages/sdk/js/src/gen/types.gen.ts +++ b/packages/sdk/js/src/gen/types.gen.ts @@ -308,6 +308,26 @@ export type StructuredOutputError = { } } +export type DoomLoopError = { + name: "DoomLoopError" + data: { + message: string + tool: string + period: number + count: number + } +} + +export type TaskBudgetExceededError = { + name: "TaskBudgetExceededError" + data: { + message: string + budget: "steps" | "tokens" | "wall_time" | "no_progress" + limit: number + used: number + } +} + export type ContextOverflowError = { name: "ContextOverflowError" data: { @@ -355,6 +375,8 @@ export type AssistantMessage = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError @@ -1420,6 +1442,8 @@ export type GlobalEvent = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError @@ -3370,18 +3394,18 @@ export type ConflictError = { resource?: string } -export type SessionNotFoundError = { - _tag: "SessionNotFoundError" - sessionID: string - message: string -} - export type ServiceUnavailableError = { _tag: "ServiceUnavailableError" message: string service?: string } +export type SessionNotFoundError = { + _tag: "SessionNotFoundError" + sessionID: string + message: string +} + export type UnknownError1 = { _tag: "UnknownError" message: string @@ -5596,6 +5620,8 @@ export type EventSessionError = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError1 @@ -6118,6 +6144,7 @@ export type GlobalCapabilitiesResponses = { 200: { protocolVersion: string version: string + commit?: string features: { im: boolean sessions: boolean @@ -7077,6 +7104,38 @@ export type DeepagentKnowledgePendingResponses = { export type DeepagentKnowledgePendingResponse = DeepagentKnowledgePendingResponses[keyof DeepagentKnowledgePendingResponses] +export type DeepagentKnowledgeReviewSummaryData = { + body?: never + path?: never + query?: { + directory?: string + workspace?: string + } + url: "/deepagent/knowledge/review-summary" +} + +export type DeepagentKnowledgeReviewSummaryErrors = { + /** + * DeepAgentPromotionError | InvalidRequestError + */ + 400: DeepAgentPromotionError | InvalidRequestError +} + +export type DeepagentKnowledgeReviewSummaryError = + DeepagentKnowledgeReviewSummaryErrors[keyof DeepagentKnowledgeReviewSummaryErrors] + +export type DeepagentKnowledgeReviewSummaryResponses = { + /** + * Pending durable knowledge count for the Review badge + */ + 200: { + pendingCount: number | "NaN" | "Infinity" | "-Infinity" | "Infinity" | "-Infinity" | "NaN" + } +} + +export type DeepagentKnowledgeReviewSummaryResponse = + DeepagentKnowledgeReviewSummaryResponses[keyof DeepagentKnowledgeReviewSummaryResponses] + export type DeepagentKnowledgeApproveData = { body?: { ids: Array @@ -14073,6 +14132,10 @@ export type V2SessionPromptErrors = { * ConflictError */ 409: ConflictError + /** + * ServiceUnavailableError + */ + 503: ServiceUnavailableError } export type V2SessionPromptError = V2SessionPromptErrors[keyof V2SessionPromptErrors] diff --git a/packages/sdk/js/src/v2/gen/sdk.gen.ts b/packages/sdk/js/src/v2/gen/sdk.gen.ts index 17ff9ae9..dc7e7a68 100644 --- a/packages/sdk/js/src/v2/gen/sdk.gen.ts +++ b/packages/sdk/js/src/v2/gen/sdk.gen.ts @@ -88,6 +88,8 @@ import type { DeepagentKnowledgeRejectIdsErrors, DeepagentKnowledgeRejectIdsResponses, DeepagentKnowledgeRejectResponses, + DeepagentKnowledgeReviewSummaryErrors, + DeepagentKnowledgeReviewSummaryResponses, DeepagentKnowledgeShipGateErrors, DeepagentKnowledgeShipGateResponses, DeepagentPacksActiveErrors, @@ -2347,6 +2349,40 @@ export class Knowledge extends HeyApiClient { }) } + /** + * Get the DeepAgent knowledge review summary + * + * Return the pending review count without projecting the complete knowledge review list. + */ + public reviewSummary( + parameters?: { + directory?: string + workspace?: string + }, + options?: Options, + ) { + const params = buildClientParams( + [parameters], + [ + { + args: [ + { in: "query", key: "directory" }, + { in: "query", key: "workspace" }, + ], + }, + ], + ) + return (options?.client ?? this.client).get< + DeepagentKnowledgeReviewSummaryResponses, + DeepagentKnowledgeReviewSummaryErrors, + ThrowOnError + >({ + url: "/deepagent/knowledge/review-summary", + ...options, + ...params, + }) + } + /** * Approve DeepAgent knowledge by id * @@ -9059,7 +9095,7 @@ export class Session3 extends HeyApiClient { /** * Send message * - * Durably admit one session input and schedule agent-loop execution unless resume is false. + * Durably admit one session input when resume is false. This endpoint does not execute the production SessionPrompt engine. */ public prompt( parameters: { diff --git a/packages/sdk/js/src/v2/gen/types.gen.ts b/packages/sdk/js/src/v2/gen/types.gen.ts index ca9bc37f..44cf0622 100644 --- a/packages/sdk/js/src/v2/gen/types.gen.ts +++ b/packages/sdk/js/src/v2/gen/types.gen.ts @@ -308,6 +308,26 @@ export type StructuredOutputError = { } } +export type DoomLoopError = { + name: "DoomLoopError" + data: { + message: string + tool: string + period: number + count: number + } +} + +export type TaskBudgetExceededError = { + name: "TaskBudgetExceededError" + data: { + message: string + budget: "steps" | "tokens" | "wall_time" | "no_progress" + limit: number + used: number + } +} + export type ContextOverflowError = { name: "ContextOverflowError" data: { @@ -355,6 +375,8 @@ export type AssistantMessage = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError @@ -1420,6 +1442,8 @@ export type GlobalEvent = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError @@ -3370,18 +3394,18 @@ export type ConflictError = { resource?: string } -export type SessionNotFoundError = { - _tag: "SessionNotFoundError" - sessionID: string - message: string -} - export type ServiceUnavailableError = { _tag: "ServiceUnavailableError" message: string service?: string } +export type SessionNotFoundError = { + _tag: "SessionNotFoundError" + sessionID: string + message: string +} + export type UnknownError1 = { _tag: "UnknownError" message: string @@ -5596,6 +5620,8 @@ export type EventSessionError = { | MessageOutputLengthError | MessageAbortedError | StructuredOutputError + | DoomLoopError + | TaskBudgetExceededError | ContextOverflowError | ApiError | OutputDegenerationError1 @@ -6118,6 +6144,7 @@ export type GlobalCapabilitiesResponses = { 200: { protocolVersion: string version: string + commit?: string features: { im: boolean sessions: boolean @@ -7077,6 +7104,38 @@ export type DeepagentKnowledgePendingResponses = { export type DeepagentKnowledgePendingResponse = DeepagentKnowledgePendingResponses[keyof DeepagentKnowledgePendingResponses] +export type DeepagentKnowledgeReviewSummaryData = { + body?: never + path?: never + query?: { + directory?: string + workspace?: string + } + url: "/deepagent/knowledge/review-summary" +} + +export type DeepagentKnowledgeReviewSummaryErrors = { + /** + * DeepAgentPromotionError | InvalidRequestError + */ + 400: DeepAgentPromotionError | InvalidRequestError +} + +export type DeepagentKnowledgeReviewSummaryError = + DeepagentKnowledgeReviewSummaryErrors[keyof DeepagentKnowledgeReviewSummaryErrors] + +export type DeepagentKnowledgeReviewSummaryResponses = { + /** + * Pending durable knowledge count for the Review badge + */ + 200: { + pendingCount: number | "NaN" | "Infinity" | "-Infinity" | "Infinity" | "-Infinity" | "NaN" + } +} + +export type DeepagentKnowledgeReviewSummaryResponse = + DeepagentKnowledgeReviewSummaryResponses[keyof DeepagentKnowledgeReviewSummaryResponses] + export type DeepagentKnowledgeApproveData = { body?: { ids: Array @@ -14073,6 +14132,10 @@ export type V2SessionPromptErrors = { * ConflictError */ 409: ConflictError + /** + * ServiceUnavailableError + */ + 503: ServiceUnavailableError } export type V2SessionPromptError = V2SessionPromptErrors[keyof V2SessionPromptErrors] From 2227c68d0c98c0264702463572f122ae02b114e3 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 07:14:16 +0800 Subject: [PATCH 18/22] fix(deepagent-code): validate tool input before execution --- packages/deepagent-code/src/session/tools.ts | 20 +++++- .../session/tool-input-validation.test.ts | 63 +++++++++++++++++++ 2 files changed, 81 insertions(+), 2 deletions(-) create mode 100644 packages/deepagent-code/test/session/tool-input-validation.test.ts diff --git a/packages/deepagent-code/src/session/tools.ts b/packages/deepagent-code/src/session/tools.ts index a6814a09..5f77762a 100644 --- a/packages/deepagent-code/src/session/tools.ts +++ b/packages/deepagent-code/src/session/tools.ts @@ -15,7 +15,8 @@ import { Plugin } from "@/plugin" import { RuntimeFlags } from "@/effect/runtime-flags" import type { TaskPromptOps } from "@/tool/task" import { type Tool as AITool, tool, jsonSchema, type ToolExecutionOptions, asSchema } from "ai" -import { Effect } from "effect" +import type { JSONSchema7 } from "@ai-sdk/provider" +import { Effect, Result, Schema } from "effect" import { MessageV2 } from "./message-v2" import { Session } from "./session" import { SessionProcessor } from "./processor" @@ -29,6 +30,21 @@ import { ToolSemanticFingerprint } from "@/tool/semantic-fingerprint" const log = Log.create({ service: "session.tools" }) +export function validatedToolInputSchema(parameters: Schema.Decoder, wireSchema: JSONSchema7) { + const decode = Schema.decodeUnknownResult(parameters) + return jsonSchema>(wireSchema, { + validate(input) { + const result = decode(input) + if (Result.isFailure(result)) { + return { success: false, error: new Error(result.failure.toString(), { cause: result.failure }) } + } + // Tool.define owns the canonical decode at execution time. Returning its decoded value here + // would apply Effect Schema transformations twice (for example NumberFromString). + return { success: true, value: input as Record } + }, + }) +} + // Plan-gate WARN reminder placement. Prepending "⚠️ Plan gate: …" AHEAD of a tool's own output was // actively harmful: the plan gate is a soft nudge — the tool ALREADY RAN and the output below the // banner is its real (usually successful, exit 0) result. But a leading "⚠️ … gate …" line reads as a @@ -196,7 +212,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: { const schema = ProviderTransform.schema(input.model, ToolJsonSchema.fromTool(item)) const aiToolDef: AITool = tool({ description: item.description, - inputSchema: jsonSchema(schema), + inputSchema: validatedToolInputSchema(item.parameters, schema), execute(args, options) { return run.promise( Effect.gen(function* () { diff --git a/packages/deepagent-code/test/session/tool-input-validation.test.ts b/packages/deepagent-code/test/session/tool-input-validation.test.ts new file mode 100644 index 00000000..24ac68ff --- /dev/null +++ b/packages/deepagent-code/test/session/tool-input-validation.test.ts @@ -0,0 +1,63 @@ +import { describe, expect, test } from "bun:test" +import { generateText, InvalidToolInputError, tool } from "ai" +import { MockLanguageModelV3 } from "ai/test" +import { Schema } from "effect" +import { SessionTools } from "../../src/session/tools" +import { ToolJsonSchema } from "../../src/tool/json-schema" +import { Parameters as Shell } from "../../src/tool/shell" + +describe("session tool input validation", () => { + test("routes missing bash arguments through AI SDK validation without executing", async () => { + const model = new MockLanguageModelV3({ + doGenerate: { + content: [{ type: "tool-call", toolCallId: "call-1", toolName: "bash", input: "{}" }], + finishReason: { unified: "tool-calls", raw: "tool_calls" }, + usage: { + inputTokens: { total: 1, noCache: 1, cacheRead: 0, cacheWrite: 0 }, + outputTokens: { total: 1, text: 1, reasoning: 0 }, + }, + warnings: [], + }, + }) + const executed: unknown[] = [] + const repairErrors: unknown[] = [] + + const result = await generateText({ + model, + prompt: "Run a command", + tools: { + bash: tool({ + inputSchema: SessionTools.validatedToolInputSchema(Shell, ToolJsonSchema.fromSchema(Shell)), + execute(input) { + executed.push(input) + return "unexpected" + }, + }), + }, + async experimental_repairToolCall(failed) { + repairErrors.push(failed.error) + return null + }, + }) + + expect(repairErrors).toHaveLength(1) + expect(InvalidToolInputError.isInstance(repairErrors[0])).toBe(true) + if (InvalidToolInputError.isInstance(repairErrors[0])) { + expect(String(repairErrors[0].cause)).toContain('["command"]') + } + expect(executed).toHaveLength(0) + expect(result.toolCalls).toMatchObject([{ toolName: "bash", invalid: true }]) + }) + + test("preserves wire input after validation so execution owns schema transformations", async () => { + const parameters = Schema.Struct({ count: Schema.NumberFromString }) + const input = { count: "7" } + const inputSchema = SessionTools.validatedToolInputSchema(parameters, ToolJsonSchema.fromSchema(parameters)) + + expect(inputSchema.validate).toBeDefined() + const result = await inputSchema.validate!(input) + + expect(result).toEqual({ success: true, value: input }) + if (result.success) expect(result.value).toBe(input) + }) +}) From 35ea93539dcc9ecef8bf304aef883ae914ef864e Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sat, 25 Jul 2026 23:54:44 +0800 Subject: [PATCH 19/22] chore: prepare 1.4.3 release --- CHANGELOG.md | 9 +++++++++ README.md | 20 +++++++++++--------- README.zh.md | 20 +++++++++++--------- bun.lock | 2 +- design/README.md | 10 ++++++++-- packages/app/README.md | 2 ++ packages/app/src/utils/solid-dnd.tsx | 1 - packages/core/package.json | 2 +- packages/deepagent-code/README.md | 2 ++ packages/desktop/README.md | 2 ++ 10 files changed, 47 insertions(+), 23 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index adb2d32a..80f9eb79 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,5 +1,14 @@ # Changelog +## Desktop 1.4.3 / DeepAgent Core V4.0.4_r8 - Production reliability release + +- Persist the complete TaskRun lifecycle, including exact-retry admission, generation, owner/lease fencing, phase changes, terminal compare-and-set, result references, and a leased notification outbox. +- Split structured subagent work into a normal research phase and a bounded single-turn finalizer with an isolated tool surface, per-turn reasoning policy, strict schema validation, and no empty-success fallback. +- Preserve distinct recoverable terminal reasons for provider, schema, permission, interruption, timeout, takeover, and doom-loop failures; reject invalid tool input before execution. +- Detect semantic no-progress from tool results, workspace snapshots, and plan state instead of relying on identical command text. +- Contain subagent supervision failures with valid sibling controls, real terminal-state rendering, a local ErrorBoundary, persisted-mode validation, same-build quarantine, Electron cold-start coverage, and production source-map verification. +- Regenerate the JavaScript SDK for the durable task status and delivery contracts. + ## V4.0.4 - Contract-gap closure (engine-decoupled) - Fix Plan Gate deadlock: stale-plan latch now warns (never hard-blocks) on tool execution, aligned with codex exec-policy philosophy. A mutating tool on a stale plan receives a reminder but always runs. diff --git a/README.md b/README.md index fb0e6be3..3db9fba3 100644 --- a/README.md +++ b/README.md @@ -14,7 +14,7 @@ Enterprise

-

Desktop 1.4.2 · DeepAgent Core V4.0.4

+

Desktop 1.4.3 · DeepAgent Core V4.0.4_r8

--- @@ -91,16 +91,18 @@ For high-risk decisions, convene an **Expert Panel**. Correctness, security, per Project IM brings people and agents into the same thread. Mention an agent to start a scoped run with project context, stream its progress, inspect its artifacts, and keep the answer attached to the conversation that requested it. -## DeepAgent Core V4.0.4 +## DeepAgent Core V4.0.4_r8 -V4.0.4 closes production contract gaps while keeping the current turn engine stable: +Desktop 1.4.3 ships the eighth reliability revision of the V4.0.4 contract. This release hardens the boundaries where long-running and delegated work previously risked duplicate execution, ambiguous completion, or renderer-wide failure: -- **Single durable truth:** DocumentStore uses atomic, recoverable writes for documents, plans, learning candidates, governance state, and version conflicts. -- **Isolated subagents:** write-capable subagents use dedicated worktrees by default and return their changes to the parent workspace through a bounded, conflict-aware path. -- **Reliable event delivery:** the Event Bus has a transport seam, durable consumer offsets, offline catch-up, real priority ordering, and observable queue depth. -- **Governed learning and goals:** knowledge promotion is tied to review evidence and ship-gate snapshots; event-driven goal ticks remain idempotent and respect quiet hours. -- **Secure integrations:** MCP credentials use environment references or native OS secret storage on macOS, Linux, and Windows; capability and source checks fail closed. -- **Publishing truth:** installation, CLI examples, release metadata, public domains, and supported-version documentation match the product that is actually shipped. +- **Durable subagent execution:** TaskRun admission, generation, ownership, leases, settlement, and parent delivery are persisted. Exact retries reconcile to the original run, terminal state uses transactional compare-and-set, and a leased outbox makes completion delivery recoverable without re-running provider work. +- **Two-stage structured finalization:** research and structured output are separate phases. The finalizer is a bounded single turn with thinking disabled when supported, only the `StructuredOutput` tool visible, no historical task or compaction payload, and no empty-success path. +- **Fail-closed model and tool boundaries:** provider capability decisions no longer silently relax required tool choice, invalid tool input is rejected before execution, and provider, schema, permission, interruption, timeout, and doom-loop failures keep distinct recoverable terminal reasons. +- **Semantic no-progress protection:** shell activity uses semantic fingerprints, while the no-progress budget compares bounded results, workspace state, and plan progress so cosmetic command changes cannot evade loop detection. +- **Contained supervision UI:** subagent controls use valid sibling interactions, display real terminal state, and run behind a local ErrorBoundary with retry, close, persisted-mode validation, same-build quarantine, and recovery after build changes. +- **Release-grade verification:** the affected Core, runtime, app, and desktop paths are covered by exact-retry, failure-injection, production Chromium, Electron cold-start, and source-map smoke tests. + +V4.0.4_r8 retains the durable documents, event delivery, governed learning, isolated worktrees, and secure credential boundaries introduced across the earlier V4.0.4 revisions. ## Installation diff --git a/README.zh.md b/README.zh.md index e6dde70a..4b1b3c38 100644 --- a/README.zh.md +++ b/README.zh.md @@ -14,7 +14,7 @@ Enterprise 版本

-

桌面版 1.4.2 · DeepAgent Core V4.0.4

+

桌面版 1.4.3 · DeepAgent Core V4.0.4_r8

--- @@ -91,16 +91,18 @@ DeepAgent 可以把独立工作拆分给数量有界、相互隔离的 Worker。 项目 IM 把团队成员和智能体放进同一条讨论。@ 某个智能体即可启动有明确作用域的运行，使用项目上下文、流式展示进度、关联执行工件，并把答案留在发起任务的对话里。 -## DeepAgent Core V4.0.4 +## DeepAgent Core V4.0.4_r8 -V4.0.4 在保持当前 turn 引擎稳定的前提下，关闭生产合同缺口： +桌面版 1.4.3 搭载 V4.0.4 合同的第八次可靠性修订。本次发布重点加固长任务和委派任务的关键边界，避免重复执行、模糊终态或局部界面故障演变为整页崩溃： -- **单一持久真相：** DocumentStore 通过原子、可恢复写入统一管理文档、计划、学习候选、治理状态和版本冲突。 -- **隔离子智能体：** 具备写权限的子智能体默认使用独立 worktree，并通过有界、可感知冲突的路径把改动回传到父工作区。 -- **可靠事件投递：** Event Bus 提供可替换 transport、持久 consumer offset、离线补投、真实优先级排序和可观测队列深度。 -- **受治理的学习与目标：** 知识晋升关联审阅证据与 ship-gate snapshot；事件驱动 goal tick 保持幂等并遵守 quiet hours。 -- **安全集成：** MCP 凭据使用环境变量引用或 macOS、Linux、Windows 原生 secret storage；capability 与来源检查逐层失败关闭。 -- **发布真实性：** 安装方式、CLI 示例、发布元数据、公开域名和支持版本文档与实际交付产品一致。 +- **持久化子智能体执行：** TaskRun 的准入、generation、owner、lease、结算与父会话投递均持久化。exact retry 会归并到原始运行，终态通过事务 CAS 结算，租约式 outbox 可在不重复执行 provider work 的前提下恢复完成通知。 +- **两阶段结构化终结：** 研究和结构化输出分为独立阶段。finalizer 是次数有界的单轮执行；在 provider 支持时关闭 thinking，只暴露 `StructuredOutput`，不混入历史 task 或 compaction 内容，也不存在空结果成功路径。 +- **失败关闭的模型与工具边界：** provider capability 决策不再静默放宽 required tool choice；无效工具输入在执行前被拒绝；provider、schema、permission、interruption、timeout 与 doom-loop 保留不同且可恢复的终态原因。 +- **语义级无进展保护：** shell 使用语义 fingerprint；无进展预算同时比较有界结果、工作区状态与计划进度，不能再靠改写命令描述绕过循环检测。 +- **故障隔离的监督面板：** 子智能体控件使用合法的 sibling interaction，展示真实终态，并由局部 ErrorBoundary、重试/关闭、持久 mode 校验、同 build quarantine 与 build 变化后的恢复机制保护。 +- **发布级验证：** 受影响的 Core、运行时、App 与 Desktop 路径已覆盖 exact retry、故障注入、production Chromium、Electron 冷启动和 source-map smoke。 + +V4.0.4_r8 同时保留此前 V4.0.4 各修订建立的持久文档、可靠事件投递、知识治理、worktree 隔离与安全凭据边界。 ## 安装 diff --git a/bun.lock b/bun.lock index 63533c10..eec42051 100644 --- a/bun.lock +++ b/bun.lock @@ -129,7 +129,7 @@ }, "packages/core": { "name": "@deepagent-code/core", - "version": "1.0.0-beta", + "version": "4.0.4-r8", "bin": { "deepagent-code": "./bin/deepagent-code", }, diff --git a/design/README.md b/design/README.md index 384f9236..07baa4b9 100644 --- a/design/README.md +++ b/design/README.md @@ -1,6 +1,6 @@ # DeepAgent Code Architecture & Design -> **Public design overview for DeepAgent Core V4.0.4 / Desktop 1.4.2.** Internal implementation details and roadmap documents live in the private `docs/` tree and are intentionally not version-controlled. +> **Public design overview for DeepAgent Core V4.0.4_r8 / Desktop 1.4.3.** Internal implementation details and roadmap documents live in the private `docs/` tree and are intentionally not version-controlled. DeepAgent Code is a document-centered, event-driven AI coding system. It combines a coding-agent runtime with a durable control plane that owns context, planning, learning, collaboration, safety, and human oversight. @@ -22,7 +22,13 @@ Sessions, inputs, plans, documents, goals, events, approvals, and learning decis A user instruction is durably admitted before execution is scheduled. A successful API response therefore means the instruction is recorded, not merely present in a process-local queue. -DeepAgent is built **on top of** the opencode agent/runtime/session/tool/MCP stack. V4.0.4 strengthens the control plane without replacing the current turn engine, tool system, or provider layer. +DeepAgent is built **on top of** the opencode agent/runtime/session/tool/MCP stack. V4.0.4_r8 strengthens the control plane without replacing the current turn engine, tool system, or provider layer. + +### Durable delegated execution + +A delegated task is admitted before provider work begins and is identified independently from the tool-call attempt that submitted it. Exact retries reconcile to the same TaskRun; conflicting reuse fails closed. Generation, execution owner, lease, phase, terminal state, result reference, and parent-notification delivery are durable records. Active mutations and terminal settlement require the expected generation, owner, and live lease, while notification delivery uses an attempt-fenced outbox so crash recovery cannot repeat provider work or acknowledge another claimant's delivery. + +Structured work uses two explicit stages. The research stage retains the child Agent's normal tools and transcript. The finalizer is a bounded, single provider turn with a narrow ephemeral tool registry, no task or compaction history, no ordinary steering, and a per-turn reasoning/tool-choice decision derived from provider capability. Completion is valid only after schema-validated structured output is persisted; provider, schema, permission, timeout, interruption, and no-progress failures remain distinct recoverable states. ### 2. One durable authority per concern diff --git a/packages/app/README.md b/packages/app/README.md index 1bf37cbd..43e278ff 100644 --- a/packages/app/README.md +++ b/packages/app/README.md @@ -2,6 +2,8 @@ SolidJS front-end shell for the DeepAgent Code desktop app (Electron/Tauri). +Current release: Desktop 1.4.3, powered by DeepAgent Core V4.0.4_r8. + ## Stack - **UI:** SolidJS + Vite (Bun) diff --git a/packages/app/src/utils/solid-dnd.tsx b/packages/app/src/utils/solid-dnd.tsx index 6c3bb3d5..b5e279e8 100644 --- a/packages/app/src/utils/solid-dnd.tsx +++ b/packages/app/src/utils/solid-dnd.tsx @@ -165,4 +165,3 @@ export const FixedDragDropSensors = (props: { children?: JSXElement }): JSXEleme createFixedPointerSensor() return props.children as JSXElement } - diff --git a/packages/core/package.json b/packages/core/package.json index dcb07b07..ed040dcb 100644 --- a/packages/core/package.json +++ b/packages/core/package.json @@ -1,6 +1,6 @@ { "$schema": "https://json.schemastore.org/package.json", - "version": "1.0.0-beta", + "version": "4.0.4-r8", "name": "@deepagent-code/core", "type": "module", "license": "AGPL-3.0-or-later", diff --git a/packages/deepagent-code/README.md b/packages/deepagent-code/README.md index 4c74998f..26b6f2d7 100644 --- a/packages/deepagent-code/README.md +++ b/packages/deepagent-code/README.md @@ -2,6 +2,8 @@ DeepAgent Code is a document-centered AI coding agent. This package contains the CLI/server runtime used by the terminal and desktop applications. +The current product release is Desktop 1.4.3 with DeepAgent Core V4.0.4_r8. + DeepAgent Code keeps the opencode runtime foundation and adds the DeepAgent control plane: - typed-document memory for run state, durable knowledge, worklogs, decisions, diagnosis, and context snapshots. diff --git a/packages/desktop/README.md b/packages/desktop/README.md index 20066dd3..3c04622a 100644 --- a/packages/desktop/README.md +++ b/packages/desktop/README.md @@ -2,6 +2,8 @@ The DeepAgent Code Desktop app, built with Electron. +Current release: Desktop 1.4.3, powered by DeepAgent Core V4.0.4_r8. + ## Development ```bash From 07e7e4057775560fe2ac057bf37a4e774d99e1a3 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sun, 26 Jul 2026 00:59:22 +0800 Subject: [PATCH 20/22] feat(provider): refresh supported model lists --- .../components/dialog-connect-provider.tsx | 20 ++--- .../components/provider-model-refresh.test.ts | 21 +++++ .../src/components/provider-model-refresh.ts | 13 +++ .../app/src/components/settings-providers.tsx | 82 +++++++++++++++---- .../src/components/settings-v2/providers.tsx | 82 +++++++++++++++---- packages/app/src/context/server-sync.tsx | 5 ++ packages/app/src/i18n/en.ts | 3 + packages/app/src/i18n/zh.ts | 3 + .../src/provider/discovery-cache.ts | 16 ++-- .../deepagent-code/src/provider/provider.ts | 70 ++++++++++++---- .../instance/httpapi/groups/provider.ts | 22 +++++ .../instance/httpapi/handlers/provider.ts | 82 ++++++++++++++++++- .../server/routes/instance/httpapi/server.ts | 4 + packages/deepagent-code/test/fake/provider.ts | 1 + .../test/provider/discovery-cache.test.ts | 13 +++ .../test/provider/provider.test.ts | 46 +++++++++++ .../test/server/httpapi-exercise/index.ts | 8 ++ packages/sdk/js/src/gen/sdk.gen.ts | 38 +++++++++ packages/sdk/js/src/gen/types.gen.ts | 34 ++++++++ packages/sdk/js/src/v2/gen/sdk.gen.ts | 38 +++++++++ packages/sdk/js/src/v2/gen/types.gen.ts | 34 ++++++++ 21 files changed, 565 insertions(+), 70 deletions(-) create mode 100644 packages/app/src/components/provider-model-refresh.test.ts create mode 100644 packages/app/src/components/provider-model-refresh.ts diff --git a/packages/app/src/components/dialog-connect-provider.tsx b/packages/app/src/components/dialog-connect-provider.tsx index 80d83dc3..8af98d18 100644 --- a/packages/app/src/components/dialog-connect-provider.tsx +++ b/packages/app/src/components/dialog-connect-provider.tsx @@ -48,11 +48,6 @@ const providerKind = (providerID: string) => (providerID === "anthropic" ? "anth type DiscoveredProviderModel = { id: string; name: string } -const discoveredModelConfig = (model: DiscoveredProviderModel) => ({ - id: model.id, - name: model.name, -}) - export function DialogConnectProvider(props: { provider: string }) { const dialog = useDialog() const serverSync = useServerSync() @@ -486,7 +481,6 @@ export function DialogConnectProvider(props: { provider: string }) { return options } - async function handleSubmit(e: SubmitEvent) { e.preventDefault() @@ -577,6 +571,8 @@ export function DialogConnectProvider(props: { provider: string }) { setFormStore("selectedModel", nextSelected.id) const current = serverSync.data.config.provider?.[props.provider] ?? {} + const keepLegacySnapshot = + current.discovery !== true && current.npm === undefined && Object.keys(current.models ?? {}).length > 0 await serverSync.updateConfig({ provider: { [props.provider]: { @@ -587,10 +583,14 @@ export function DialogConnectProvider(props: { provider: string }) { baseURL, ...(key ? { apiKey: key } : {}), }, - models: { - ...(current.models ?? {}), - ...Object.fromEntries(nextModels.map((model) => [model.id, discoveredModelConfig(model)])), - }, + // Keep the provider's live /models endpoint authoritative after the initial import. + // Existing discovery-mode entries may contain intentional per-model overrides. Legacy + // static snapshots stay untagged because config updates merge nested maps and cannot + // safely clear them; the backend recognizes that old shape and ignores the snapshot + // whenever a live list is available. + ...(keepLegacySnapshot + ? {} + : { discovery: true, models: current.discovery ? (current.models ?? {}) : {} }), }, }, disabled_providers: (serverSync.data.config.disabled_providers ?? []).filter((id) => id !== props.provider), diff --git a/packages/app/src/components/provider-model-refresh.test.ts b/packages/app/src/components/provider-model-refresh.test.ts new file mode 100644 index 00000000..20fb6526 --- /dev/null +++ b/packages/app/src/components/provider-model-refresh.test.ts @@ -0,0 +1,21 @@ +import { describe, expect, test } from "bun:test" +import { canRefreshProviderModels } from "./provider-model-refresh" + +describe("provider model refresh", () => { + test("allows official, discovery, and legacy imported providers", () => { + expect(canRefreshProviderModels("openai", undefined)).toBe(true) + expect(canRefreshProviderModels("custom", { discovery: true })).toBe(true) + expect( + canRefreshProviderModels("mistral", { + options: { baseURL: "https://api.mistral.ai/v1" }, + models: { mistral: { name: "Mistral" } }, + }), + ).toBe(true) + expect( + canRefreshProviderModels("manual", { + npm: "@ai-sdk/openai-compatible", + options: { baseURL: "https://manual.example/v1" }, + }), + ).toBe(false) + }) +}) diff --git a/packages/app/src/components/provider-model-refresh.ts b/packages/app/src/components/provider-model-refresh.ts new file mode 100644 index 00000000..90786742 --- /dev/null +++ b/packages/app/src/components/provider-model-refresh.ts @@ -0,0 +1,13 @@ +import { isOfficialProvider } from "@deepagent-code/core/provider-official" +import type { ProviderConfig } from "@deepagent-code/sdk/v2" + +export function canRefreshProviderModels(providerID: string, config: ProviderConfig | undefined) { + if (isOfficialProvider(providerID)) return true + if (config?.discovery === true) return true + return ( + config?.npm === undefined && + typeof config?.options?.baseURL === "string" && + !!config.options.baseURL && + Object.keys(config.models ?? {}).length > 0 + ) +} diff --git a/packages/app/src/components/settings-providers.tsx b/packages/app/src/components/settings-providers.tsx index 855668c1..07f5da31 100644 --- a/packages/app/src/components/settings-providers.tsx +++ b/packages/app/src/components/settings-providers.tsx @@ -5,6 +5,7 @@ import { Tag } from "@deepagent-code/ui/tag" import { showToast } from "@/utils/toast" import { popularProviders, useProviders } from "@/hooks/use-providers" import { createMemo, type Component, For, Show } from "solid-js" +import { createStore } from "solid-js/store" import { useLanguage } from "@/context/language" import { useServerSDK } from "@/context/server-sdk" import { useServerSync } from "@/context/server-sync" @@ -13,6 +14,7 @@ import { DialogSelectProvider } from "./dialog-select-provider" import { DialogCustomProvider } from "./dialog-custom-provider" import { SettingsList } from "./settings-list" import { SettingsServerPicker, SettingsServerScope } from "./settings-server-picker" +import { canRefreshProviderModels } from "./provider-model-refresh" type ProviderSource = "env" | "api" | "config" | "custom" type ProviderItem = ReturnType["connected"]>[number] @@ -46,6 +48,7 @@ const SettingsProvidersContent: Component = () => { const serverSDK = useServerSDK() const serverSync = useServerSync() const providers = useProviders() + const [refreshing, setRefreshing] = createStore>({}) const isConfigCustom = (providerID: string) => { const provider = serverSync.data.config.provider?.[providerID] @@ -92,6 +95,32 @@ const SettingsProvidersContent: Component = () => { const note = (id: string) => PROVIDER_NOTES.find((item) => item.match(id))?.key + const canRefresh = (providerID: string) => + canRefreshProviderModels(providerID, serverSync.data.config.provider?.[providerID]) + + const refreshModels = async (providerID: string, name: string) => { + if (refreshing[providerID]) return + setRefreshing(providerID, true) + await serverSDK.client.provider.models + .refresh({ providerID }, { throwOnError: true }) + .then((result) => { + serverSync.refreshProviders() + showToast({ + variant: "success", + icon: "circle-check", + title: language.t("provider.models.refresh.toast.title", { provider: name }), + description: language.t("provider.models.refresh.toast.description", { + count: Object.keys(result.data?.models ?? {}).length, + }), + }) + }) + .catch((err: unknown) => { + const message = err instanceof Error ? err.message : String(err) + showToast({ title: language.t("common.requestFailed"), description: message }) + }) + .finally(() => setRefreshing(providerID, false)) + } + const disableProvider = async (providerID: string, name: string) => { const before = serverSync.data.config.disabled_providers ?? [] const next = before.includes(providerID) ? before : [...before, providerID] @@ -180,25 +209,42 @@ const SettingsProvidersContent: Component = () => { {item.name} {type(item)} - - {language.t("settings.providers.connected.environmentDescription")} - - } - > - + + + {language.t("settings.providers.connected.environmentDescription")} + + } > - {language.t("common.disconnect")} - - + + + )} diff --git a/packages/app/src/components/settings-v2/providers.tsx b/packages/app/src/components/settings-v2/providers.tsx index 85b596e5..7baaff52 100644 --- a/packages/app/src/components/settings-v2/providers.tsx +++ b/packages/app/src/components/settings-v2/providers.tsx @@ -5,6 +5,7 @@ import { ProviderIcon } from "@deepagent-code/ui/provider-icon" import { showToast } from "@/utils/toast" import { popularProviders, useProviders } from "@/hooks/use-providers" import { createMemo, type Component, For, Show } from "solid-js" +import { createStore } from "solid-js/store" import { useLanguage } from "@/context/language" import { useServerSDK } from "@/context/server-sdk" import { useServerSync } from "@/context/server-sync" @@ -12,6 +13,7 @@ import { DialogConnectProvider } from "../dialog-connect-provider" import { DialogSelectProvider } from "../dialog-select-provider" import { DialogCustomProvider } from "../dialog-custom-provider" import { SettingsListV2 } from "./parts/list" +import { canRefreshProviderModels } from "../provider-model-refresh" import "./settings-v2.css" type ProviderSource = "env" | "api" | "config" | "custom" @@ -40,6 +42,7 @@ export const SettingsProvidersV2: Component = () => { const serverSdk = useServerSDK() const serverSync = useServerSync() const providers = useProviders() + const [refreshing, setRefreshing] = createStore>({}) const isConfigCustom = (providerID: string) => { const provider = serverSync.data.config.provider?.[providerID] @@ -91,6 +94,32 @@ export const SettingsProvidersV2: Component = () => { const note = (id: string) => PROVIDER_NOTES.find((item) => item.match(id))?.key + const canRefresh = (providerID: string) => + canRefreshProviderModels(providerID, serverSync.data.config.provider?.[providerID]) + + const refreshModels = async (providerID: string, name: string) => { + if (refreshing[providerID]) return + setRefreshing(providerID, true) + await serverSdk.client.provider.models + .refresh({ providerID }, { throwOnError: true }) + .then((result) => { + serverSync.refreshProviders() + showToast({ + variant: "success", + icon: "circle-check", + title: language.t("provider.models.refresh.toast.title", { provider: name }), + description: language.t("provider.models.refresh.toast.description", { + count: Object.keys(result.data?.models ?? {}).length, + }), + }) + }) + .catch((err: unknown) => { + const message = err instanceof Error ? err.message : String(err) + showToast({ title: language.t("common.requestFailed"), description: message }) + }) + .finally(() => setRefreshing(providerID, false)) + } + const disableProvider = async (providerID: string, name: string) => { const before = serverSync.data.config.disabled_providers ?? [] const next = before.includes(providerID) ? before : [...before, providerID] @@ -235,25 +264,42 @@ export const SettingsProvidersV2: Component = () => { {type(item)} - - {language.t("settings.providers.connected.environmentDescription")} - - } - > - { - event.stopPropagation() - void disconnect(item.id, item.name) - }} +
+ + { + event.stopPropagation() + void refreshModels(item.id, item.name) + }} + > + {refreshing[item.id] + ? language.t("provider.models.refresh.refreshing") + : language.t("common.refresh")} + + + + {language.t("settings.providers.connected.environmentDescription")} + + } > - {language.t("common.disconnect")} - - + { + event.stopPropagation() + void disconnect(item.id, item.name) + }} + > + {language.t("common.disconnect")} + + +
)} diff --git a/packages/app/src/context/server-sync.tsx b/packages/app/src/context/server-sync.tsx index 91f79cc5..dfd38d15 100644 --- a/packages/app/src/context/server-sync.tsx +++ b/packages/app/src/context/server-sync.tsx @@ -50,6 +50,8 @@ import type { SessionPlan, SessionPlanStep, SessionGoal } from "./global-sync/ty export type { SessionPlan, SessionPlanStep, SessionGoal } +const PROVIDER_MODEL_REFRESH_INTERVAL_MS = 24 * 60 * 60 * 1000 + type GlobalStore = { ready: boolean error?: InitError @@ -547,6 +549,9 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) { }) } + const providerRefreshTimer = setInterval(refreshProviders, PROVIDER_MODEL_REFRESH_INTERVAL_MS) + onCleanup(() => clearInterval(providerRefreshTimer)) + const updateConfigMutation = useMutation(() => ({ mutationFn: (config: Config) => serverSDK.client.global.config.update({ config }), onSuccess: () => { diff --git a/packages/app/src/i18n/en.ts b/packages/app/src/i18n/en.ts index 0f1bd014..b385af78 100644 --- a/packages/app/src/i18n/en.ts +++ b/packages/app/src/i18n/en.ts @@ -176,6 +176,9 @@ export const dict = { "provider.connect.oauth.auto.confirmationCode": "Confirmation code", "provider.connect.toast.connected.title": "{{provider}} connected", "provider.connect.toast.connected.description": "{{provider}} models are now available to use.", + "provider.models.refresh.refreshing": "Refreshing...", + "provider.models.refresh.toast.title": "{{provider}} models refreshed", + "provider.models.refresh.toast.description": "{{count}} models are now available.", "provider.connect.transport.title": "Advanced: request timeouts & retries", "provider.connect.transport.description": diff --git a/packages/app/src/i18n/zh.ts b/packages/app/src/i18n/zh.ts index def41e6d..f9c26a02 100644 --- a/packages/app/src/i18n/zh.ts +++ b/packages/app/src/i18n/zh.ts @@ -201,6 +201,9 @@ export const dict = { "provider.connect.oauth.auto.confirmationCode": "确认码", "provider.connect.toast.connected.title": "{{provider}} 已连接", "provider.connect.toast.connected.description": "现在可以使用 {{provider}} 模型了。", + "provider.models.refresh.refreshing": "刷新中...", + "provider.models.refresh.toast.title": "{{provider}} 模型已刷新", + "provider.models.refresh.toast.description": "当前共有 {{count}} 个可用模型。", "provider.connect.transport.title": "高级：请求超时与重试", "provider.connect.transport.description": "官方 provider 的传输设置独立于配置文件保存。留空则使用默认值。", diff --git a/packages/deepagent-code/src/provider/discovery-cache.ts b/packages/deepagent-code/src/provider/discovery-cache.ts index 2df55385..0a5384c2 100644 --- a/packages/deepagent-code/src/provider/discovery-cache.ts +++ b/packages/deepagent-code/src/provider/discovery-cache.ts @@ -5,7 +5,12 @@ import { Hash } from "@deepagent-code/core/util/hash" import { Log } from "@deepagent-code/core/util/log" import type { FSUtil } from "@deepagent-code/core/fs-util" import type { EffectFlock } from "@deepagent-code/core/util/effect-flock" -import { discoverProviderModels, isChatModel, type DiscoveredModel, type ProviderDiscoveryKind } from "./model-discovery" +import { + discoverProviderModels, + isChatModel, + type DiscoveredModel, + type ProviderDiscoveryKind, +} from "./model-discovery" const log = Log.create({ service: "provider-discovery-cache" }) @@ -76,6 +81,7 @@ export const discoverModelsCached = Effect.fn("ProviderDiscovery.cached")(functi flock: EffectFlock.Interface, input: DiscoverModelsCachedInput, fetch: (input: DiscoverModelsCachedInput) => Promise = discoverProviderModels, + force = false, ) { const ttl = input.ttl ?? DEFAULT_DISCOVERY_TTL const filepath = cacheFile(input) @@ -94,7 +100,7 @@ export const discoverModelsCached = Effect.fn("ProviderDiscovery.cached")(functi return Date.now() - mtime < Duration.toMillis(ttl) }) - if (yield* fresh) { + if (!force && (yield* fresh)) { const cached = yield* readDisk if (cached) return cached } @@ -110,9 +116,7 @@ export const discoverModelsCached = Effect.fn("ProviderDiscovery.cached")(functi yield* fs.writeWithDirs(tempfile, JSON.stringify(models)).pipe( Effect.andThen(fs.rename(tempfile, filepath)), Effect.catch((error) => - fs - .remove(tempfile, { force: true }) - .pipe(Effect.ignore, Effect.andThen(Effect.fail(error))), + fs.remove(tempfile, { force: true }).pipe(Effect.ignore, Effect.andThen(Effect.fail(error))), ), ) return models @@ -123,7 +127,7 @@ export const discoverModelsCached = Effect.fn("ProviderDiscovery.cached")(functi return yield* flock .withLock( Effect.gen(function* () { - if (yield* fresh) { + if (!force && (yield* fresh)) { const cached = yield* readDisk if (cached) return cached } diff --git a/packages/deepagent-code/src/provider/provider.ts b/packages/deepagent-code/src/provider/provider.ts index 4049c528..f12e5caa 100644 --- a/packages/deepagent-code/src/provider/provider.ts +++ b/packages/deepagent-code/src/provider/provider.ts @@ -59,6 +59,7 @@ const LEGACY_AUTH_KEY_MESSAGE = "Saved API key is no longer used. Only official providers read keys from the key store. Re-add this as a third-party provider in your config and set the key under options.apiKey." const DISCOVERY_EMPTY_MESSAGE = "Runtime model discovery returned no models. Check the base URL, API key, and that the endpoint implements GET /models, or list models explicitly under provider..models." +const MODEL_LIST_REFRESH_INTERVAL_MS = 24 * 60 * 60 * 1000 function providerEnvKey(configuredEnv: string[], envs: Record) { const configured = configuredEnv.find((item) => envs[item]) @@ -1109,6 +1110,7 @@ export type Error = ModelNotFoundError | InitError | NoProvidersError | NoModels export interface Interface { readonly list: () => Effect.Effect> readonly errors: () => Effect.Effect + readonly reload: () => Effect.Effect readonly getProvider: (providerID: ProviderV2.ID) => Effect.Effect readonly getModel: (providerID: ProviderV2.ID, modelID: ModelV2.ID) => Effect.Effect readonly getLanguage: (model: Model) => Effect.Effect @@ -1121,6 +1123,7 @@ export interface Interface { } interface State { + loadedAt: number models: Map providers: Record catalog: Record @@ -1444,10 +1447,18 @@ export const layer = Layer.effect( // Track providers that opted into discovery so the zero-model deletion below can tell a // discovery provider that came up empty (worth an error) apart from a normal empty provider. const discoveryProviders = new Set() + // Older connect flows persisted the first /models response as an untagged static snapshot. + // A catalog provider with no explicit npm is the stable signature of that flow; treat it as + // discovery-enabled so existing users gain refresh without rewriting or losing config. + const legacyDiscoveryProviders = new Set() yield* Effect.forEach( configProviders.filter( ([providerID, provider]) => - provider.discovery === true && + (provider.discovery === true || + (provider.npm === undefined && + typeof provider.options?.baseURL === "string" && + Object.keys(provider.models ?? {}).length > 0 && + modelsDev[providerID] !== undefined)) && // Both reserved hosted ids (the product gateway "deepagent-code" and the routed // "deepagent") own their catalog identity — never run third-party discovery against them. providerID !== "deepagent-code" && @@ -1458,6 +1469,7 @@ export const layer = Layer.effect( ([providerID, provider]) => Effect.gen(function* () { discoveryProviders.add(providerID) + if (provider.discovery !== true) legacyDiscoveryProviders.add(providerID) const baseURL = provider.options?.baseURL if (!baseURL) return const envKey = provider.env ? providerEnvKey(provider.env, envs) : undefined @@ -1465,7 +1477,8 @@ export const layer = Layer.effect( const headers = provider.options?.headers as Record | undefined // A key OR custom auth headers must be present; a bare baseURL can't authenticate. if (!apiKey && !headers) return - const kind: ProviderDiscoveryKind = provider.npm === "@ai-sdk/anthropic" ? "anthropic" : "openai-compatible" + const kind: ProviderDiscoveryKind = + (provider.npm ?? modelsDev[providerID]?.npm) === "@ai-sdk/anthropic" ? "anthropic" : "openai-compatible" const models = yield* discoverModelsCached(fs, flock, { providerID, baseURL, apiKey, kind, headers }) if (!models.length) return discoveredModels[providerID] = Object.fromEntries( @@ -1480,7 +1493,8 @@ export const layer = Layer.effect( const gApiKey = group.options?.apiKey ?? apiKey const gHeaders = { ...headers, ...(group.options?.headers as Record | undefined) } if (!gBaseURL || (!gApiKey && !Object.keys(gHeaders).length)) continue - const gKind: ProviderDiscoveryKind = group.npm === "@ai-sdk/anthropic" ? "anthropic" : "openai-compatible" + const gKind: ProviderDiscoveryKind = + group.npm === "@ai-sdk/anthropic" ? "anthropic" : "openai-compatible" const groupCacheKey = `${providerID}:${groupId}` const gModels = yield* discoverModelsCached(fs, flock, { providerID: groupCacheKey, @@ -1534,7 +1548,10 @@ export const layer = Layer.effect( // Discovered models seed the source map; hand-listed `provider.models` override them so an // explicit config entry always wins over the runtime-discovered version of the same id. - const modelSource = { ...(discoveredModels[providerID] ?? {}), ...(provider.models ?? {}) } + const modelSource = + legacyDiscoveryProviders.has(providerID) && discoveredModels[providerID] + ? discoveredModels[providerID] + : { ...(discoveredModels[providerID] ?? {}), ...(provider.models ?? {}) } for (const [modelID, model] of Object.entries(modelSource)) { const existingModel = parsed.models[model.id ?? modelID] const apiID = model.id ?? existingModel?.api.id ?? modelID @@ -1958,6 +1975,7 @@ export const layer = Layer.effect( } return { + loadedAt: Date.now(), models: languages, providers, catalog, @@ -1969,8 +1987,20 @@ export const layer = Layer.effect( }), ) - const list = Effect.fn("Provider.list")(() => InstanceState.use(state, (s) => s.providers)) - const errors = Effect.fn("Provider.errors")(() => InstanceState.use(state, (s) => s.errors)) + const getState = Effect.fnUntraced(function* () { + const current = yield* InstanceState.get(state) + if (Date.now() - current.loadedAt < MODEL_LIST_REFRESH_INTERVAL_MS) return current + yield* InstanceState.invalidate(state) + return yield* InstanceState.get(state) + }) + + const list = Effect.fn("Provider.list")(function* () { + return (yield* getState()).providers + }) + const errors = Effect.fn("Provider.errors")(function* () { + return (yield* getState()).errors + }) + const reload = Effect.fn("Provider.reload")(() => InstanceState.invalidate(state)) function deepagentModelAuthProviderID(model: Model) { if (model.providerID !== "deepagent") return @@ -2126,12 +2156,12 @@ export const layer = Layer.effect( } } - const getProvider = Effect.fn("Provider.getProvider")((providerID: ProviderV2.ID) => - InstanceState.use(state, (s) => s.providers[providerID]), - ) + const getProvider = Effect.fn("Provider.getProvider")(function* (providerID: ProviderV2.ID) { + return (yield* getState()).providers[providerID] + }) const getModel = Effect.fn("Provider.getModel")(function* (providerID: ProviderV2.ID, modelID: ModelV2.ID) { - const s = yield* InstanceState.get(state) + const s = yield* getState() const provider = s.providers[providerID] if (!provider) { const catalogProvider = s.catalog[providerID] @@ -2155,7 +2185,7 @@ export const layer = Layer.effect( }) const getLanguage = Effect.fn("Provider.getLanguage")(function* (model: Model) { - const s = yield* InstanceState.get(state) + const s = yield* getState() const envs = yield* env.all() const key = `${model.providerID}/${model.id}` if (s.models.has(key)) return s.models.get(key)! @@ -2188,7 +2218,7 @@ export const layer = Layer.effect( }) const closest = Effect.fn("Provider.closest")(function* (providerID: ProviderV2.ID, query: string[]) { - const s = yield* InstanceState.get(state) + const s = yield* getState() const provider = s.providers[providerID] if (!provider) return undefined for (const item of query) { @@ -2209,7 +2239,7 @@ export const layer = Layer.effect( ) } - const s = yield* InstanceState.get(state) + const s = yield* getState() const provider = s.providers[providerID] if (!provider) return undefined @@ -2273,7 +2303,7 @@ export const layer = Layer.effect( const cfg = yield* config.get() if (cfg.model) return parseModel(cfg.model) - const s = yield* InstanceState.get(state) + const s = yield* getState() const recent = yield* fs.readJson(path.join(Global.Path.state, "model.json")).pipe( Effect.map((x): { providerID: ProviderV2.ID; modelID: ModelV2.ID }[] => { if (!isRecord(x) || !Array.isArray(x.recent)) return [] @@ -2303,7 +2333,17 @@ export const layer = Layer.effect( } }) - return Service.of({ list, errors, getProvider, getModel, getLanguage, closest, getSmallModel, defaultModel }) + return Service.of({ + list, + errors, + reload, + getProvider, + getModel, + getLanguage, + closest, + getSmallModel, + defaultModel, + }) }), ) diff --git a/packages/deepagent-code/src/server/routes/instance/httpapi/groups/provider.ts b/packages/deepagent-code/src/server/routes/instance/httpapi/groups/provider.ts index d48dd365..0d434034 100644 --- a/packages/deepagent-code/src/server/routes/instance/httpapi/groups/provider.ts +++ b/packages/deepagent-code/src/server/routes/instance/httpapi/groups/provider.ts @@ -40,6 +40,15 @@ export class ProviderModelDiscoverError extends Schema.ErrorClass( + "ProviderModelRefreshError", +)( + { + message: Schema.String, + }, + { httpApiStatus: 400 }, +) {} + export const ProviderModelDiscoverInput = Schema.Struct({ providerID: Schema.String, baseURL: Schema.String, @@ -115,6 +124,19 @@ export const ProviderApi = HttpApi.make("provider") description: "Probe a provider /models endpoint and return discovered chat models.", }), ), + HttpApiEndpoint.post("refreshModels", `${root}/:providerID/models/refresh`, { + params: { providerID: ProviderV2.ID }, + query: WorkspaceRoutingQuery, + success: described(Provider.Info, "Provider with refreshed models"), + error: ProviderModelRefreshError, + }).annotateMerge( + OpenApi.annotations({ + identifier: "provider.models.refresh", + summary: "Refresh provider models", + description: + "Force-refresh one provider's supported model list and rebuild the provider registry for this instance.", + }), + ), HttpApiEndpoint.post("authorize", `${root}/:providerID/oauth/authorize`, { params: { providerID: ProviderV2.ID }, query: WorkspaceRoutingQuery, diff --git a/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/provider.ts b/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/provider.ts index cb16b00e..c6336c1a 100644 --- a/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/provider.ts +++ b/packages/deepagent-code/src/server/routes/instance/httpapi/handlers/provider.ts @@ -1,18 +1,22 @@ import { ProviderAuth } from "@/provider/auth" import { Auth } from "@/auth" import { Config } from "@/config/config" +import { Env } from "@/env" import { ModelsDev } from "@deepagent-code/core/models-dev" import { Provider } from "@/provider/provider" -import { discoverWithProtocol, isChatModel, normalizeBaseURL } from "@/provider/model-discovery" +import { discoverProviderModels, discoverWithProtocol, isChatModel, normalizeBaseURL } from "@/provider/model-discovery" +import { discoverModelsCached } from "@/provider/discovery-cache" import { buildCatalogIndex, projectSpec, specMatchFor } from "@/provider/catalog-spec" +import { FSUtil } from "@deepagent-code/core/fs-util" +import { EffectFlock } from "@deepagent-code/core/util/effect-flock" import { mapValues } from "remeda" import { Effect, Schema } from "effect" import { HttpServerRequest, HttpServerResponse } from "effect/unstable/http" import { HttpApiBuilder } from "effect/unstable/httpapi" import { InstanceHttpApi } from "../api" -import { ProviderAuthApiError, ProviderModelDiscoverError } from "../groups/provider" -import { ProviderV2 } from "@deepagent-code/core/provider" +import { ProviderAuthApiError, ProviderModelDiscoverError, ProviderModelRefreshError } from "../groups/provider" +import { OFFICIAL_PROVIDER_ID_SET, ProviderV2 } from "@deepagent-code/core/provider" function mapProviderAuthError(self: Effect.Effect) { return self.pipe( @@ -40,6 +44,10 @@ export const providerHandlers = HttpApiBuilder.group(InstanceHttpApi, "provider" const provider = yield* Provider.Service const svc = yield* ProviderAuth.Service const authSvc = yield* Auth.Service + const env = yield* Env.Service + const modelsDev = yield* ModelsDev.Service + const fs = yield* FSUtil.Service + const flock = yield* EffectFlock.Service const list = Effect.fn("ProviderHttpApi.list")(function* () { const config = yield* cfg.get() @@ -124,6 +132,73 @@ export const providerHandlers = HttpApiBuilder.group(InstanceHttpApi, "provider" return { providerID, baseURL, kind: result.kind, models: selectable, selected } }) + const refreshModels = Effect.fn("ProviderHttpApi.refreshModels")(function* (ctx: { + params: { providerID: ProviderV2.ID } + }) { + const providerID = ctx.params.providerID + if (OFFICIAL_PROVIDER_ID_SET.has(providerID)) { + yield* modelsDev.refresh(true) + yield* provider.reload() + const refreshed = yield* provider.getProvider(providerID) + if (refreshed) return Provider.toPublicInfo(refreshed) + return yield* new ProviderModelRefreshError({ message: `Provider not found: ${providerID}` }) + } + + const item = (yield* cfg.get()).provider?.[providerID] + const catalog = yield* modelsDev.get() + const legacyDiscovery = + item?.npm === undefined && + typeof item?.options?.baseURL === "string" && + Object.keys(item.models ?? {}).length > 0 && + catalog[providerID] !== undefined + if (!item || (!item.discovery && !legacyDiscovery)) { + return yield* new ProviderModelRefreshError({ + message: `Provider ${providerID} does not have runtime model discovery enabled`, + }) + } + + const baseURL = item.options?.baseURL + if (typeof baseURL !== "string" || !baseURL.trim()) { + return yield* new ProviderModelRefreshError({ message: `Provider ${providerID} is missing a base URL` }) + } + + const envs = yield* env.all() + const apiKey = + typeof item.options?.apiKey === "string" + ? item.options.apiKey + : item.env?.map((key) => envs[key]).find((value): value is string => typeof value === "string" && !!value) + const headers = + item.options?.headers && typeof item.options.headers === "object" + ? Object.fromEntries( + Object.entries(item.options.headers).filter( + (entry): entry is [string, string] => typeof entry[1] === "string", + ), + ) + : undefined + if (!apiKey && !headers) { + return yield* new ProviderModelRefreshError({ + message: `Provider ${providerID} is missing discovery credentials`, + }) + } + + const kind = (item.npm ?? catalog[providerID]?.npm) === "@ai-sdk/anthropic" ? "anthropic" : "openai-compatible" + const input = { providerID, baseURL, apiKey, kind, headers } as const + const discovered = yield* Effect.tryPromise({ + try: () => discoverProviderModels(input), + catch: (error) => + new ProviderModelRefreshError({ message: error instanceof Error ? error.message : String(error) }), + }) + const models = yield* discoverModelsCached(fs, flock, input, () => Promise.resolve(discovered), true) + if (!models.length) { + return yield* new ProviderModelRefreshError({ message: `Provider ${providerID} returned no chat models` }) + } + + yield* provider.reload() + const refreshed = yield* provider.getProvider(providerID) + if (refreshed) return Provider.toPublicInfo(refreshed) + return yield* new ProviderModelRefreshError({ message: `Provider not found after refresh: ${providerID}` }) + }) + const authorize = Effect.fn("ProviderHttpApi.authorize")(function* (ctx: { params: { providerID: ProviderV2.ID } payload: ProviderAuth.AuthorizeInput @@ -170,6 +245,7 @@ export const providerHandlers = HttpApiBuilder.group(InstanceHttpApi, "provider" .handle("list", list) .handle("auth", auth) .handle("discover", discover) + .handle("refreshModels", refreshModels) .handleRaw("authorize", authorizeRaw) .handle("callback", callback) }), diff --git a/packages/deepagent-code/src/server/routes/instance/httpapi/server.ts b/packages/deepagent-code/src/server/routes/instance/httpapi/server.ts index 7b3e0e37..a10d67d6 100644 --- a/packages/deepagent-code/src/server/routes/instance/httpapi/server.ts +++ b/packages/deepagent-code/src/server/routes/instance/httpapi/server.ts @@ -10,6 +10,7 @@ import { } from "effect/unstable/http" import * as Socket from "effect/unstable/socket/Socket" import { FSUtil } from "@deepagent-code/core/fs-util" +import { EffectFlock } from "@deepagent-code/core/util/effect-flock" import { FileLock } from "@deepagent-code/core/file-lock" import { Account } from "@/account/account" import { Agent } from "@/agent/agent" @@ -17,6 +18,7 @@ import { Auth } from "@/auth" import { BackgroundJob } from "@/background/job" import { Config } from "@/config/config" import { Command } from "@/command" +import { Env } from "@/env" import * as Observability from "@deepagent-code/core/effect/observability" import { Ripgrep } from "@deepagent-code/core/filesystem/ripgrep" import { Format } from "@/format" @@ -330,6 +332,7 @@ export function createRoutes( BackgroundJob.defaultLayer, Command.defaultLayer, Config.defaultLayer, + Env.defaultLayer, Format.defaultLayer, LSP.defaultLayer, LLM.defaultLayer, @@ -373,6 +376,7 @@ export function createRoutes( Workspace.defaultLayer, Worktree.appLayer, FSUtil.defaultLayer, + EffectFlock.defaultLayer, FileLock.layer, FetchHttpClient.layer, HttpServer.layerServices, diff --git a/packages/deepagent-code/test/fake/provider.ts b/packages/deepagent-code/test/fake/provider.ts index 70cac03c..2a681a7d 100644 --- a/packages/deepagent-code/test/fake/provider.ts +++ b/packages/deepagent-code/test/fake/provider.ts @@ -55,6 +55,7 @@ export namespace ProviderTest { Provider.Service.of({ list: Effect.fn("TestProvider.list")(() => Effect.succeed({ [row.id]: row })), errors: Effect.fn("TestProvider.errors")(() => Effect.succeed([])), + reload: Effect.fn("TestProvider.reload")(() => Effect.void), getProvider: Effect.fn("TestProvider.getProvider")((providerID) => { if (providerID === row.id) return Effect.succeed(row) return Effect.die(new Error(`Unknown test provider: ${providerID}`)) diff --git a/packages/deepagent-code/test/provider/discovery-cache.test.ts b/packages/deepagent-code/test/provider/discovery-cache.test.ts index f5827409..8c170cae 100644 --- a/packages/deepagent-code/test/provider/discovery-cache.test.ts +++ b/packages/deepagent-code/test/provider/discovery-cache.test.ts @@ -80,6 +80,19 @@ describe("discoverModelsCached", () => { expect(calls).toBe(2) }) + test("force refresh bypasses a fresh cache", async () => { + let calls = 0 + const fetch = async () => [model(`forced-${++calls}`)] + const input = baseInput("forced") + + const first = await run((fs, flock) => discoverModelsCached(fs, flock, input, fetch)) + expect(first.map((m) => m.id)).toEqual(["forced-1"]) + + const second = await run((fs, flock) => discoverModelsCached(fs, flock, input, fetch, true)) + expect(second.map((m) => m.id)).toEqual(["forced-2"]) + expect(calls).toBe(2) + }) + test("falls back to the last good cache when a refetch fails", async () => { let calls = 0 const fetch = async () => { diff --git a/packages/deepagent-code/test/provider/provider.test.ts b/packages/deepagent-code/test/provider/provider.test.ts index 4a273c7b..4326a738 100644 --- a/packages/deepagent-code/test/provider/provider.test.ts +++ b/packages/deepagent-code/test/provider/provider.test.ts @@ -57,6 +57,7 @@ const remove = (k: string) => }) afterEach(async () => { + discoveryGeneration = 1 for (const [key, value] of originalEnv) { if (value === undefined) delete process.env[key] else process.env[key] = value @@ -115,6 +116,7 @@ const connect = (providerID: ProviderV2.ID, key: string) => let discoveryServer: ReturnType | undefined let discoveryServerURL = "" let discoveryEmptyURL = "" +let discoveryGeneration = 1 beforeAll(() => { discoveryServer = Bun.serve({ @@ -125,6 +127,11 @@ beforeAll(() => { // `/empty/models` returns a 200 with no models (provisioning / not-implemented shape); every // other path returns two chat models plus one embedding model that runtime filtering must drop. if (url.pathname.startsWith("/empty")) return Response.json({ data: [] }) + if (url.pathname.startsWith("/refresh")) { + return Response.json({ + data: [{ id: `runtime-${discoveryGeneration}`, display_name: `Runtime ${discoveryGeneration}` }], + }) + } return Response.json({ data: [ { id: "runtime-a", display_name: "Runtime A" }, @@ -239,6 +246,45 @@ it.instance( }, ) +it.instance( + "legacy imported provider snapshots refresh automatically after one day", + Effect.gen(function* () { + const providerID = ProviderV2.ID.openrouter + const first = (yield* list)[providerID] + expect(first.models["runtime-1"]).toBeDefined() + expect(first.models["legacy-snapshot"]).toBeUndefined() + + discoveryGeneration = 2 + const now = Date.now + yield* Effect.acquireUseRelease( + Effect.sync(() => { + Date.now = () => now() + 25 * 60 * 60 * 1000 + }), + () => + Effect.gen(function* () { + const refreshed = (yield* list)[providerID] + expect(refreshed.models["runtime-1"]).toBeUndefined() + expect(refreshed.models["runtime-2"]).toBeDefined() + }), + () => + Effect.sync(() => { + Date.now = now + }), + ) + }), + { + config: () => ({ + provider: { + openrouter: { + name: "OpenRouter", + options: { apiKey: "k", baseURL: `${discoveryServerURL}/../refresh` }, + models: { "legacy-snapshot": { name: "Legacy Snapshot" } }, + }, + }, + }), + }, +) + it.instance( "manual models win over discovered models for the same id", Effect.gen(function* () { diff --git a/packages/deepagent-code/test/server/httpapi-exercise/index.ts b/packages/deepagent-code/test/server/httpapi-exercise/index.ts index 1544c589..dfb8420d 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/index.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/index.ts @@ -268,6 +268,14 @@ const scenarios: Scenario[] = [ .status(204, undefined, "status"), http.protected.get("/provider", "provider.list").json(), http.protected.get("/provider/auth", "provider.auth").json(), + http.protected + .post("/provider/{providerID}/models/refresh", "provider.models.refresh") + .probe({ path: "/provider/models/discover", body: {} }) + .at((ctx) => ({ + path: route("/provider/{providerID}/models/refresh", { providerID: "httpapi" }), + headers: ctx.headers(), + })) + .status(400), http.protected .post("/provider/{providerID}/oauth/authorize", "provider.oauth.authorize") .at((ctx) => ({ diff --git a/packages/sdk/js/src/gen/sdk.gen.ts b/packages/sdk/js/src/gen/sdk.gen.ts index dc7e7a68..0dffd653 100644 --- a/packages/sdk/js/src/gen/sdk.gen.ts +++ b/packages/sdk/js/src/gen/sdk.gen.ts @@ -339,6 +339,8 @@ import type { ProviderListResponses, ProviderModelsDiscoverErrors, ProviderModelsDiscoverResponses, + ProviderModelsRefreshErrors, + ProviderModelsRefreshResponses, ProviderOauthAuthorizeErrors, ProviderOauthAuthorizeResponses, ProviderOauthCallbackErrors, @@ -6766,6 +6768,42 @@ export class Models extends HeyApiClient { }, }) } + + /** + * Refresh provider models + * + * Force-refresh one provider's supported model list and rebuild the provider registry for this instance. + */ + public refresh( + parameters: { + providerID: string + directory?: string + workspace?: string + }, + options?: Options, + ) { + const params = buildClientParams( + [parameters], + [ + { + args: [ + { in: "path", key: "providerID" }, + { in: "query", key: "directory" }, + { in: "query", key: "workspace" }, + ], + }, + ], + ) + return (options?.client ?? this.client).post< + ProviderModelsRefreshResponses, + ProviderModelsRefreshErrors, + ThrowOnError + >({ + url: "/provider/{providerID}/models/refresh", + ...options, + ...params, + }) + } } export class Oauth extends HeyApiClient { diff --git a/packages/sdk/js/src/gen/types.gen.ts b/packages/sdk/js/src/gen/types.gen.ts index 44cf0622..ed2ba90e 100644 --- a/packages/sdk/js/src/gen/types.gen.ts +++ b/packages/sdk/js/src/gen/types.gen.ts @@ -3184,6 +3184,10 @@ export type ProviderModelDiscoverError = { message: string } +export type ProviderModelRefreshError = { + message: string +} + export type ProviderAuthAuthorization = { url: string method: "auto" | "code" @@ -11881,6 +11885,36 @@ export type ProviderModelsDiscoverResponses = { export type ProviderModelsDiscoverResponse = ProviderModelsDiscoverResponses[keyof ProviderModelsDiscoverResponses] +export type ProviderModelsRefreshData = { + body?: never + path: { + providerID: string + } + query?: { + directory?: string + workspace?: string + } + url: "/provider/{providerID}/models/refresh" +} + +export type ProviderModelsRefreshErrors = { + /** + * ProviderModelRefreshError | InvalidRequestError + */ + 400: ProviderModelRefreshError | InvalidRequestError +} + +export type ProviderModelsRefreshError = ProviderModelsRefreshErrors[keyof ProviderModelsRefreshErrors] + +export type ProviderModelsRefreshResponses = { + /** + * Provider with refreshed models + */ + 200: Provider +} + +export type ProviderModelsRefreshResponse = ProviderModelsRefreshResponses[keyof ProviderModelsRefreshResponses] + export type ProviderOauthAuthorizeData = { body?: { /** diff --git a/packages/sdk/js/src/v2/gen/sdk.gen.ts b/packages/sdk/js/src/v2/gen/sdk.gen.ts index dc7e7a68..0dffd653 100644 --- a/packages/sdk/js/src/v2/gen/sdk.gen.ts +++ b/packages/sdk/js/src/v2/gen/sdk.gen.ts @@ -339,6 +339,8 @@ import type { ProviderListResponses, ProviderModelsDiscoverErrors, ProviderModelsDiscoverResponses, + ProviderModelsRefreshErrors, + ProviderModelsRefreshResponses, ProviderOauthAuthorizeErrors, ProviderOauthAuthorizeResponses, ProviderOauthCallbackErrors, @@ -6766,6 +6768,42 @@ export class Models extends HeyApiClient { }, }) } + + /** + * Refresh provider models + * + * Force-refresh one provider's supported model list and rebuild the provider registry for this instance. + */ + public refresh( + parameters: { + providerID: string + directory?: string + workspace?: string + }, + options?: Options, + ) { + const params = buildClientParams( + [parameters], + [ + { + args: [ + { in: "path", key: "providerID" }, + { in: "query", key: "directory" }, + { in: "query", key: "workspace" }, + ], + }, + ], + ) + return (options?.client ?? this.client).post< + ProviderModelsRefreshResponses, + ProviderModelsRefreshErrors, + ThrowOnError + >({ + url: "/provider/{providerID}/models/refresh", + ...options, + ...params, + }) + } } export class Oauth extends HeyApiClient { diff --git a/packages/sdk/js/src/v2/gen/types.gen.ts b/packages/sdk/js/src/v2/gen/types.gen.ts index 44cf0622..ed2ba90e 100644 --- a/packages/sdk/js/src/v2/gen/types.gen.ts +++ b/packages/sdk/js/src/v2/gen/types.gen.ts @@ -3184,6 +3184,10 @@ export type ProviderModelDiscoverError = { message: string } +export type ProviderModelRefreshError = { + message: string +} + export type ProviderAuthAuthorization = { url: string method: "auto" | "code" @@ -11881,6 +11885,36 @@ export type ProviderModelsDiscoverResponses = { export type ProviderModelsDiscoverResponse = ProviderModelsDiscoverResponses[keyof ProviderModelsDiscoverResponses] +export type ProviderModelsRefreshData = { + body?: never + path: { + providerID: string + } + query?: { + directory?: string + workspace?: string + } + url: "/provider/{providerID}/models/refresh" +} + +export type ProviderModelsRefreshErrors = { + /** + * ProviderModelRefreshError | InvalidRequestError + */ + 400: ProviderModelRefreshError | InvalidRequestError +} + +export type ProviderModelsRefreshError = ProviderModelsRefreshErrors[keyof ProviderModelsRefreshErrors] + +export type ProviderModelsRefreshResponses = { + /** + * Provider with refreshed models + */ + 200: Provider +} + +export type ProviderModelsRefreshResponse = ProviderModelsRefreshResponses[keyof ProviderModelsRefreshResponses] + export type ProviderOauthAuthorizeData = { body?: { /** From 80212a2a9b3986ccbbc8c11c20bc927831959109 Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sun, 26 Jul 2026 01:43:40 +0800 Subject: [PATCH 21/22] fix(deepagent-code): isolate http api exerciser state --- .../deepagent-code/script/httpapi-exercise.ts | 1 + .../httpapi-exercise/environment-bootstrap.ts | 29 +++++++++++++++++++ .../server/httpapi-exercise/environment.ts | 26 ++++++----------- 3 files changed, 39 insertions(+), 17 deletions(-) create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/environment-bootstrap.ts diff --git a/packages/deepagent-code/script/httpapi-exercise.ts b/packages/deepagent-code/script/httpapi-exercise.ts index 5395a812..f270be2e 100644 --- a/packages/deepagent-code/script/httpapi-exercise.ts +++ b/packages/deepagent-code/script/httpapi-exercise.ts @@ -1 +1,2 @@ +await import("../test/server/httpapi-exercise/environment-bootstrap") await import("../test/server/httpapi-exercise/index") diff --git a/packages/deepagent-code/test/server/httpapi-exercise/environment-bootstrap.ts b/packages/deepagent-code/test/server/httpapi-exercise/environment-bootstrap.ts new file mode 100644 index 00000000..2431c2f8 --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/environment-bootstrap.ts @@ -0,0 +1,29 @@ +import path from "path" + +export const preserveExerciseGlobalRoot = !!process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_GLOBAL +export const exerciseGlobalRoot = + process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_GLOBAL ?? + path.join(process.env.TMPDIR ?? "/tmp", `deepagent-code-httpapi-global-${process.pid}`) + +process.env.XDG_DATA_HOME = path.join(exerciseGlobalRoot, "data") +process.env.XDG_CONFIG_HOME = path.join(exerciseGlobalRoot, "config") +process.env.XDG_STATE_HOME = path.join(exerciseGlobalRoot, "state") +process.env.XDG_CACHE_HOME = path.join(exerciseGlobalRoot, "cache") +process.env.DEEPAGENT_CODE_DISABLE_SHARE = "true" + +export const exerciseConfigDirectory = path.join(exerciseGlobalRoot, "config", "deepagent-code") +export const exerciseDataDirectory = path.join(exerciseGlobalRoot, "data", "deepagent-code") +process.env.DEEPAGENT_CODE_HOME = exerciseDataDirectory +process.env.DEEPAGENT_CODE_EXPERIMENTAL_EXPERT_PANEL = "false" +process.env.DEEPAGENT_CODE_EXPERIMENTAL_GOAL_LOOP = "false" +process.env.DEEPAGENT_CODE_EXPERIMENTAL_WIKI = "false" +process.env.DEEPAGENT_CODE_V4_FILE_UPLOAD_ENABLED = "false" +process.env.DEEPAGENT_CODE_V4_MULTI_AGENT_RUNTIME = "false" +process.env.DEEPAGENT_CODE_V4_THREAD_ENABLED = "false" + +export const preserveExerciseDatabase = !!process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_DB +export const exerciseDatabasePath = + process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_DB ?? + path.join(process.env.TMPDIR ?? "/tmp", `deepagent-code-httpapi-exercise-${process.pid}.db`) + +process.env.DEEPAGENT_CODE_DB = exerciseDatabasePath diff --git a/packages/deepagent-code/test/server/httpapi-exercise/environment.ts b/packages/deepagent-code/test/server/httpapi-exercise/environment.ts index 03dcac4d..c7cf7840 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/environment.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/environment.ts @@ -1,24 +1,16 @@ import { Flag } from "@deepagent-code/core/flag/flag" import { Effect } from "effect" -import path from "path" +import { + exerciseConfigDirectory, + exerciseDataDirectory, + exerciseDatabasePath, + exerciseGlobalRoot, + preserveExerciseDatabase, + preserveExerciseGlobalRoot, +} from "./environment-bootstrap" -const preserveExerciseGlobalRoot = !!process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_GLOBAL -export const exerciseGlobalRoot = - process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_GLOBAL ?? - path.join(process.env.TMPDIR ?? "/tmp", `deepagent-code-httpapi-global-${process.pid}`) -process.env.XDG_DATA_HOME = path.join(exerciseGlobalRoot, "data") -process.env.XDG_CONFIG_HOME = path.join(exerciseGlobalRoot, "config") -process.env.XDG_STATE_HOME = path.join(exerciseGlobalRoot, "state") -process.env.XDG_CACHE_HOME = path.join(exerciseGlobalRoot, "cache") -process.env.DEEPAGENT_CODE_DISABLE_SHARE = "true" -export const exerciseConfigDirectory = path.join(exerciseGlobalRoot, "config", "deepagent-code") -export const exerciseDataDirectory = path.join(exerciseGlobalRoot, "data", "deepagent-code") +export { exerciseConfigDirectory, exerciseDataDirectory, exerciseDatabasePath, exerciseGlobalRoot } -const preserveExerciseDatabase = !!process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_DB -export const exerciseDatabasePath = - process.env.DEEPAGENT_CODE_HTTPAPI_EXERCISE_DB ?? - path.join(process.env.TMPDIR ?? "/tmp", `deepagent-code-httpapi-exercise-${process.pid}.db`) -process.env.DEEPAGENT_CODE_DB = exerciseDatabasePath Flag.DEEPAGENT_CODE_DB = exerciseDatabasePath export const original = { From dabfffc4868691e8250c05c450faaacdd63d49ad Mon Sep 17 00:00:00 2001 From: deepagent-ai Date: Sun, 26 Jul 2026 01:43:49 +0800 Subject: [PATCH 22/22] test(deepagent-code): cover all http api routes --- .../test/server/httpapi-exercise/backend.ts | 12 +- .../test/server/httpapi-exercise/dsl.ts | 4 + .../test/server/httpapi-exercise/index.ts | 12 + .../httpapi-exercise/scenarios/deepagent.ts | 225 ++++++++++++++++++ .../server/httpapi-exercise/scenarios/file.ts | 175 ++++++++++++++ .../server/httpapi-exercise/scenarios/im.ts | 133 +++++++++++ .../httpapi-exercise/scenarios/oversight.ts | 57 +++++ .../httpapi-exercise/scenarios/platform.ts | 161 +++++++++++++ .../httpapi-exercise/scenarios/runtime.ts | 112 +++++++++ .../test/server/httpapi-exercise/types.ts | 2 +- 10 files changed, 891 insertions(+), 2 deletions(-) create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/deepagent.ts create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/file.ts create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/im.ts create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/oversight.ts create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/platform.ts create mode 100644 packages/deepagent-code/test/server/httpapi-exercise/scenarios/runtime.ts diff --git a/packages/deepagent-code/test/server/httpapi-exercise/backend.ts b/packages/deepagent-code/test/server/httpapi-exercise/backend.ts index cf8b9ff4..eacfbe3e 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/backend.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/backend.ts @@ -118,7 +118,17 @@ function authProbePath(path: string) { } async function capture(response: Response, mode: CaptureMode): Promise { - const text = mode === "stream" ? await captureStream(response) : await response.text() + const text = + mode === "stream" + ? await captureStream(response) + : mode === "headers" + ? response.body + ? await response.body.cancel().then( + () => "", + () => "", + ) + : "" + : await response.text() return { status: response.status, contentType: response.headers.get("content-type") ?? "", diff --git a/packages/deepagent-code/test/server/httpapi-exercise/dsl.ts b/packages/deepagent-code/test/server/httpapi-exercise/dsl.ts index 97ff92f1..8bda0516 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/dsl.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/dsl.ts @@ -66,6 +66,10 @@ class ScenarioBuilder { return this.clone({ capture: "stream" }) } + headersOnly() { + return this.clone({ capture: "headers" }) + } + protected() { return this.auth("protected") } diff --git a/packages/deepagent-code/test/server/httpapi-exercise/index.ts b/packages/deepagent-code/test/server/httpapi-exercise/index.ts index dfb8420d..2e5bc44f 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/index.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/index.ts @@ -37,6 +37,12 @@ import { runScenario } from "./runner" import { disposeApps } from "./backend" import { runtime } from "./runtime" import { type Scenario } from "./types" +import { platformScenarios } from "./scenarios/platform" +import { fileScenarios } from "./scenarios/file" +import { imScenarios } from "./scenarios/im" +import { runtimeScenarios } from "./scenarios/runtime" +import { deepagentScenarios } from "./scenarios/deepagent" +import { oversightScenarios } from "./scenarios/oversight" void (await import("@deepagent-code/core/util/log")).init({ print: false }) @@ -75,6 +81,12 @@ const deepagentCandidate = (id: string) => ({ }) const scenarios: Scenario[] = [ + ...platformScenarios, + ...fileScenarios, + ...imScenarios, + ...runtimeScenarios, + ...deepagentScenarios, + ...oversightScenarios, http.protected .get("/global/health", "global.health") .global() diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/deepagent.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/deepagent.ts new file mode 100644 index 00000000..4ca26796 --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/deepagent.ts @@ -0,0 +1,225 @@ +import { AgentGateway } from "@deepagent-code/core/agent-gateway" +import { Effect } from "effect" +import path from "path" +import { array, check, object } from "../assertions" +import { http } from "../dsl" +import { exerciseDataDirectory } from "../environment" +import type { Scenario } from "../types" + +AgentGateway.configure({ enabled: true, runsDir: path.join(exerciseDataDirectory, "deepagent", "runs") }) + +export const deepagentScenarios: Scenario[] = [ + http.protected.get("/deepagent/knowledge/pending", "deepagent.knowledge.pending").json(200, (body) => { + object(body) + array(body.items) + }), + http.protected.get("/deepagent/knowledge/review-summary", "deepagent.knowledge.reviewSummary").json(200, (body) => { + object(body) + check(typeof body.pendingCount === "number", "knowledge review summary should return a count") + }), + http.protected + .post("/deepagent/knowledge/approve", "deepagent.knowledge.approve") + .mutating() + .at((ctx) => ({ path: "/deepagent/knowledge/approve", headers: ctx.headers(), body: { ids: [] } })) + .json(200, (body) => { + object(body) + check(Array.isArray(body.updated) && body.updated.length === 0, "empty knowledge approval should be idempotent") + }), + http.protected + .post("/deepagent/knowledge/reject-ids", "deepagent.knowledge.rejectIds") + .mutating() + .at((ctx) => ({ path: "/deepagent/knowledge/reject-ids", headers: ctx.headers(), body: { ids: [] } })) + .json(200, (body) => { + object(body) + check(Array.isArray(body.updated) && body.updated.length === 0, "empty knowledge rejection should be idempotent") + }), + http.protected + .post("/deepagent/knowledge/ship-gate", "deepagent.knowledge.shipGate") + .at((ctx) => ({ + path: "/deepagent/knowledge/ship-gate", + headers: ctx.headers(), + body: { + tasks: ["httpapi"], + metrics: [ + { group: "general", task: "httpapi", metric: 1 }, + { group: "high", task: "httpapi", metric: 1 }, + { group: "max", task: "httpapi", metric: 1 }, + ], + candidateRefs: [], + }, + })) + .json(200, (body) => { + object(body) + check(body.ship === true, "equal measured metrics should pass the knowledge ship gate") + array(body.offenders) + object(body.per_group) + }), + http.protected.get("/deepagent/packs/active", "deepagent.packs.active").json(200, (body) => { + object(body) + array(body.packs) + check(typeof body.snapshotId === "string", "active packs should carry a deterministic snapshot id") + }), + http.protected.get("/deepagent/packs/all", "deepagent.packs.all").json(200, (body) => { + object(body) + array(body.packs) + }), + http.protected + .post("/deepagent/packs/pin", "deepagent.packs.pin") + .mutating() + .at((ctx) => ({ path: "/deepagent/packs/pin", headers: ctx.headers(), body: { packId: "httpapi-pack" } })) + .json(200, (body) => { + object(body) + check(body.ok === true && body.packId === "httpapi-pack", "pack pin should persist the requested id") + }), + http.protected + .post("/deepagent/packs/unpin", "deepagent.packs.unpin") + .mutating() + .at((ctx) => ({ path: "/deepagent/packs/unpin", headers: ctx.headers(), body: { packId: "httpapi-pack" } })) + .json(200, (body) => { + object(body) + check(body.ok === true && body.packId === "httpapi-pack", "pack unpin should be idempotent") + }), + http.protected.get("/deepagent/env-facts", "deepagent.envFacts.list").json(200, (body) => { + object(body) + array(body.adopted) + array(body.pending) + }), + http.protected + .post("/deepagent/env-facts/decide", "deepagent.envFacts.decide") + .mutating() + .at((ctx) => ({ + path: "/deepagent/env-facts/decide", + headers: ctx.headers(), + body: { factId: "fact_httpapi", decision: "reject" }, + })) + .json(200, (body) => { + object(body) + check(body.ok === true && body.factId === "fact_httpapi", "environment fact decision should be recorded") + }), + http.protected + .post("/deepagent/env-facts/modify", "deepagent.envFacts.modify") + .mutating() + .at((ctx) => ({ + path: "/deepagent/env-facts/modify", + headers: ctx.headers(), + body: { + factId: "fact_httpapi", + description: "HTTP API environment fact", + body: { host: "localhost", port: 3000, last_confirmed_at: "2026-01-01T00:00:00.000Z" }, + mode: "project", + }, + })) + .json(200, (body) => { + object(body) + check( + body.ok === true && typeof body.factId === "string", + "environment fact modification should return the stored id", + ) + }), + http.protected + .post("/deepagent/panel/consult", "deepagent.panel.consult") + .at((ctx) => ({ + path: "/deepagent/panel/consult", + headers: ctx.headers(), + body: { sessionID: "session_httpapi_panel", question: "Review the HTTP API" }, + })) + .json(400, object, "status"), + http.protected + .post("/deepagent/panel/arm", "deepagent.panel.arm") + .mutating() + .seeded(() => Effect.sync(() => AgentGateway.DeepAgentSessionState.getOrCreate("session_httpapi_panel", "high"))) + .at((ctx) => ({ + path: "/deepagent/panel/arm", + headers: ctx.headers(), + body: { sessionID: "session_httpapi_panel", armed: true, rounds: "multi" }, + })) + .json(200, (body) => { + object(body) + check(body.armed === true && body.rounds === "multi", "panel arm should persist the requested mode") + }), + http.protected + .get("/deepagent/panel/status", "deepagent.panel.status") + .at((ctx) => ({ path: "/deepagent/panel/status?sessionID=session_httpapi_status", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + check(body.sessionID === "session_httpapi_status", "panel status should echo the session id") + check( + typeof body.armed === "boolean" && typeof body.explicit === "boolean", + "panel status should expose its source", + ) + }), + http.protected + .post("/deepagent/goal/start", "deepagent.goal.start") + .at((ctx) => ({ + path: "/deepagent/goal/start", + headers: ctx.headers(), + body: { sessionID: "session_httpapi_goal", objective: "Exercise the route" }, + })) + .json(400, object, "status"), + ...["pause", "resume", "stop"].map((action) => + http.protected + .post(`/deepagent/goal/${action}`, `deepagent.goal.${action}`) + .at((ctx) => ({ + path: `/deepagent/goal/${action}`, + headers: ctx.headers(), + body: { sessionID: "session_httpapi_goal" }, + })) + .json(200, (body) => { + object(body) + check(body.ok === false, `disabled goal ${action} should fail closed without mutation`) + }), + ), + http.protected + .post("/deepagent/goal/edit-plan", "deepagent.goal.editPlan") + .at((ctx) => ({ + path: "/deepagent/goal/edit-plan", + headers: ctx.headers(), + body: { + sessionID: "session_httpapi_goal", + plan: { goal: "Exercise the route", steps: [{ title: "Verify the response" }] }, + }, + })) + .json(200, (body) => { + object(body) + check(body.ok === false, "disabled goal plan editing should fail closed") + }), + http.protected + .get("/deepagent/goal/status", "deepagent.goal.status") + .at((ctx) => ({ path: "/deepagent/goal/status?sessionID=session_httpapi_goal", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + check(body.goal === null, "an inactive session should have no goal") + }), + http.protected + .get("/deepagent/goal/startable", "deepagent.goal.startable") + .at((ctx) => ({ path: "/deepagent/goal/startable?sessionID=session_httpapi_goal", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + check(typeof body.startable === "boolean", "goal startability should be explicit") + check(["plan", "file", "none"].includes(String(body.source)), "goal startability should identify its source") + }), + http.protected.get("/deepagent/wiki/pages", "deepagent.wiki.pages").json(400, object, "status"), + http.protected + .get("/deepagent/wiki/page", "deepagent.wiki.page") + .at((ctx) => ({ path: "/deepagent/wiki/page?docId=doc_httpapi&scope=durable", headers: ctx.headers() })) + .json(400, object, "status"), + http.protected + .get("/deepagent/wiki/search", "deepagent.wiki.search") + .at((ctx) => ({ path: "/deepagent/wiki/search?text=httpapi", headers: ctx.headers() })) + .json(400, object, "status"), + http.protected + .post("/deepagent/wiki/edit", "deepagent.wiki.edit") + .at((ctx) => ({ + path: "/deepagent/wiki/edit", + headers: ctx.headers(), + body: { docId: "doc_httpapi", scope: "durable", body: "updated", editor: { id: "httpapi" } }, + })) + .json(400, object, "status"), + http.protected + .get("/deepagent/wiki/execution-archive", "deepagent.wiki.executionArchive") + .at((ctx) => ({ + path: "/deepagent/wiki/execution-archive?sessionID=session_httpapi_archive", + headers: ctx.headers(), + })) + .json(400, object, "status"), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/file.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/file.ts new file mode 100644 index 00000000..44adb28e --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/file.ts @@ -0,0 +1,175 @@ +import { Effect } from "effect" +import path from "path" +import { array, check, isRecord, object } from "../assertions" +import { http } from "../dsl" +import type { Scenario } from "../types" + +export const fileScenarios: Scenario[] = [ + http.protected + .post("/file/create", "file.create") + .mutating() + .at((ctx) => ({ + path: "/file/create", + headers: ctx.headers(), + body: { path: "created.txt", content: "created\n" }, + })) + .jsonEffect(200, (body, ctx) => + Effect.gen(function* () { + object(body) + check(body.ok === true && body.path === "created.txt", "file create should report the created path") + check( + (yield* Effect.promise(() => Bun.file(path.join(ctx.directory!, "created.txt")).text())) === "created\n", + "file create should persist the requested content", + ) + }), + ), + http.protected + .post("/file/delete", "file.delete") + .mutating() + .seeded((ctx) => ctx.file("delete.txt", "delete me\n")) + .at((ctx) => ({ path: "/file/delete", headers: ctx.headers(), body: { path: "delete.txt" } })) + .jsonEffect(200, (body, ctx) => + Effect.gen(function* () { + object(body) + check(body.ok === true, "file delete should report success") + check( + !(yield* Effect.promise(() => Bun.file(path.join(ctx.directory!, "delete.txt")).exists())), + "file delete should remove the target", + ) + }), + ), + http.protected + .post("/file/mkdir", "file.mkdir") + .mutating() + .at((ctx) => ({ path: "/file/mkdir", headers: ctx.headers(), body: { path: "nested/directory" } })) + .jsonEffect(200, (body, ctx) => + Effect.gen(function* () { + object(body) + check(body.ok === true, "mkdir should report success") + const fs = yield* Effect.promise(() => import("fs/promises")) + check( + (yield* Effect.promise(() => fs.stat(path.join(ctx.directory!, "nested/directory")))).isDirectory(), + "mkdir should create the directory", + ) + }), + ), + http.protected + .post("/file/rename", "file.rename") + .mutating() + .seeded((ctx) => ctx.file("before.txt", "renamed\n")) + .at((ctx) => ({ + path: "/file/rename", + headers: ctx.headers(), + body: { from: "before.txt", to: "after.txt" }, + })) + .jsonEffect(200, (body, ctx) => + Effect.gen(function* () { + object(body) + check(body.ok === true && body.path === "after.txt", "rename should report the destination") + check( + (yield* Effect.promise(() => Bun.file(path.join(ctx.directory!, "after.txt")).text())) === "renamed\n", + "rename should preserve file content", + ) + }), + ), + http.protected + .post("/file/write", "file.write") + .mutating() + .seeded((ctx) => ctx.file("write.txt", "before\n")) + .at((ctx) => ({ + path: "/file/write", + headers: ctx.headers(), + body: { path: "write.txt", content: "after\n" }, + })) + .jsonEffect(200, (body, ctx) => + Effect.gen(function* () { + object(body) + check(body.ok === true, "file write should report success") + check( + (yield* Effect.promise(() => Bun.file(path.join(ctx.directory!, "write.txt")).text())) === "after\n", + "file write should replace the target content", + ) + }), + ), + http.protected + .post("/file/lock", "file.lock.acquire") + .at((ctx) => ({ path: "/file/lock", headers: ctx.headers(), body: { path: "locked.txt", kind: "human" } })) + .json(200, (body) => { + object(body) + check(body.ok === true && isRecord(body.lock), "lock acquire should return a lock") + check(isRecord(body.lock) && body.lock.kind === "human", "lock acquire should preserve the owner kind") + }), + http.protected + .post("/file/lock/renew", "file.lock.renew") + .at((ctx) => ({ path: "/file/lock/renew", headers: ctx.headers(), body: { lockId: "lock_httpapi_missing" } })) + .json(200, (body) => { + object(body) + check(body.ok === false, "renewing a missing lock should fail closed") + }), + http.protected + .post("/file/lock/release", "file.lock.release") + .at((ctx) => ({ path: "/file/lock/release", headers: ctx.headers(), body: { lockId: "lock_httpapi_missing" } })) + .json(200, (body) => { + object(body) + check(body.ok === true, "releasing a missing lock should remain idempotent") + }), + http.protected + .get("/file/lock/status", "file.lock.status") + .at((ctx) => ({ path: "/file/lock/status?path=unlocked.txt", headers: ctx.headers() })) + .json(200, (body) => check(body === null, "an unlocked path should return null")), + http.protected + .get("/lsp/diagnostics", "lsp.diagnostics") + .seeded((ctx) => ctx.file("diagnostics.ts", "export const value = 1\n")) + .at((ctx) => ({ path: "/lsp/diagnostics?path=diagnostics.ts", headers: ctx.headers() })) + .json(200, object), + http.protected + .post("/lsp/hover", "lsp.hover") + .seeded((ctx) => ctx.file("hover.ts", "export const value = 1\n")) + .at((ctx) => ({ path: "/lsp/hover", headers: ctx.headers(), body: { file: "hover.ts", line: 0, character: 13 } })) + .json(), + http.protected + .post("/lsp/definition", "lsp.definition") + .seeded((ctx) => ctx.file("definition.ts", "export const value = 1\n")) + .at((ctx) => ({ + path: "/lsp/definition", + headers: ctx.headers(), + body: { file: "definition.ts", line: 0, character: 13 }, + })) + .json(), + http.protected + .post("/lsp/completion", "lsp.completion") + .seeded((ctx) => ctx.file("completion.ts", "const value = Math.\n")) + .at((ctx) => ({ + path: "/lsp/completion", + headers: ctx.headers(), + body: { file: "completion.ts", line: 0, character: 19 }, + })) + .json(), + http.protected + .post("/lsp/code-action", "lsp.codeAction") + .seeded((ctx) => ctx.file("action.ts", "export const value = 1\n")) + .at((ctx) => ({ + path: "/lsp/code-action", + headers: ctx.headers(), + body: { file: "action.ts", startLine: 0, startCharacter: 0, endLine: 0, endCharacter: 22 }, + })) + .json(200, array), + http.protected + .post("/lsp/rename", "lsp.rename") + .seeded((ctx) => ctx.file("rename.ts", "export const value = 1\n")) + .at((ctx) => ({ + path: "/lsp/rename", + headers: ctx.headers(), + body: { file: "rename.ts", line: 0, character: 13, newName: "nextValue" }, + })) + .json(), + http.protected.get("/mcp/catalog", "mcp.catalog").json(200, array), + http.protected + .post("/mcp/catalog/enable", "mcp.catalogEnable") + .at((ctx) => ({ + path: "/mcp/catalog/enable", + headers: ctx.headers(), + body: { id: "httpapi-missing", params: {}, credentialRefs: {} }, + })) + .json(404, object, "status"), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/im.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/im.ts new file mode 100644 index 00000000..5d8dedf2 --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/im.ts @@ -0,0 +1,133 @@ +import { array, check, object } from "../assertions" +import { http, route } from "../dsl" +import type { Scenario } from "../types" + +export const imScenarios: Scenario[] = [ + http.protected.get("/api/v1/im/groups", "im.groups.list").json(200, array), + http.protected + .post("/api/v1/im/groups", "im.groups.create") + .mutating() + .at((ctx) => ({ + path: "/api/v1/im/groups", + headers: ctx.headers(), + body: { name: "HTTP API Group", type: "project" }, + })) + .json(200, (body) => { + object(body) + check(body.name === "HTTP API Group", "created IM group should preserve its name") + check(body.type === "project", "created IM group should preserve its type") + }), + http.protected + .get("/api/v1/im/groups/{groupId}/messages", "im.messages.list") + .at((ctx) => ({ + path: route("/api/v1/im/groups/{groupId}/messages", { groupId: "group_httpapi_missing" }), + headers: ctx.headers(), + })) + .json(404, object, "status"), + http.protected + .post("/api/v1/im/groups/{groupId}/messages", "im.messages.create") + .at((ctx) => ({ + path: route("/api/v1/im/groups/{groupId}/messages", { groupId: "group_httpapi_missing" }), + headers: ctx.headers(), + body: { senderType: "user", type: "text", content: "HTTP API message" }, + })) + .json(404, object, "status"), + http.protected + .post("/api/v1/im/groups/{groupId}/read", "im.messages.markRead") + .at((ctx) => ({ + path: route("/api/v1/im/groups/{groupId}/read", { groupId: "group_httpapi_missing" }), + headers: ctx.headers(), + body: {}, + })) + .json(404, object, "status"), + http.protected.get("/api/v1/im/agents", "im.agents.list").json(200, array), + http.protected + .get("/api/v1/im/messages/{messageId}", "im.messages.get") + .at((ctx) => ({ + path: route("/api/v1/im/messages/{messageId}", { messageId: "message_httpapi_missing" }), + headers: ctx.headers(), + })) + .json(404, object, "status"), + http.protected + .get("/api/v1/im/groups/{groupId}/messages/{messageId}/thread", "im.messages.thread") + .at((ctx) => ({ + path: route("/api/v1/im/groups/{groupId}/messages/{messageId}/thread", { + groupId: "group_httpapi_missing", + messageId: "message_httpapi_missing", + }), + headers: ctx.headers(), + })) + .json(404, object, "status"), + http.protected + .get("/api/v1/im/search", "im.messages.search") + .at((ctx) => ({ path: "/api/v1/im/search?q=httpapi", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + array(body.messages) + check(typeof body.hasMore === "boolean", "IM search should return pagination metadata") + }), + http.protected + .post("/api/v1/im/attachments", "im.attachments.upload") + .probe({ path: "/api/v1/im/attachments", body: {} }) + .at((ctx) => ({ path: "/api/v1/im/attachments", headers: ctx.headers(), body: {} })) + .status(415), + http.protected.get("/api/v1/im/attachments", "im.attachments.list").json(404, object, "status"), + http.protected + .get("/ws/im/group/{groupId}", "im.websocket.connect") + .at((ctx) => ({ + path: route("/ws/im/group/{groupId}", { groupId: "group_httpapi_missing" }), + headers: ctx.headers(), + })) + .status(404, undefined, "status"), + http.protected + .post("/api/v1/webhook/git", "webhook.git") + .mutating() + .at((ctx) => ({ + path: "/api/v1/webhook/git", + headers: ctx.headers(), + body: { repo: "deepagent-code", commit: "httpapi-git", deliveryId: "httpapi-git" }, + })) + .json(200, (body) => { + object(body) + check(body.accepted === true && body.type === "git.push", "git webhook should publish a git.push event") + }), + http.protected + .post("/api/v1/webhook/ci", "webhook.ci") + .mutating() + .at((ctx) => ({ + path: "/api/v1/webhook/ci", + headers: ctx.headers(), + body: { repo: "deepagent-code", pipeline: "httpapi", deliveryId: "httpapi-ci" }, + })) + .json(200, (body) => { + object(body) + check(body.accepted === true && body.type === "ci.failure", "CI webhook should publish a ci.failure event") + }), + http.protected + .post("/api/v1/webhook/pr", "webhook.pr") + .mutating() + .at((ctx) => ({ + path: "/api/v1/webhook/pr", + headers: ctx.headers(), + body: { repo: "deepagent-code", comment: "Please review", deliveryId: "httpapi-pr" }, + })) + .json(200, (body) => { + object(body) + check(body.accepted === true && body.type === "pr.comment", "PR webhook should publish a pr.comment event") + }), + http.protected + .post("/api/v1/webhook/monitor", "webhook.monitor") + .mutating() + .at((ctx) => ({ + path: "/api/v1/webhook/monitor", + headers: ctx.headers(), + body: { title: "HTTP API alert", severity: "warning", deliveryId: "httpapi-monitor" }, + })) + .json(200, (body) => { + object(body) + check( + body.accepted === true && body.type === "monitor.alert", + "monitor webhook should publish a monitor.alert event", + ) + }), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/oversight.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/oversight.ts new file mode 100644 index 00000000..4a15b31c --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/oversight.ts @@ -0,0 +1,57 @@ +import { array, check, object } from "../assertions" +import { http } from "../dsl" +import type { Scenario } from "../types" + +export const oversightScenarios: Scenario[] = [ + http.protected.get("/oversight/metrics", "oversight.metrics").json(200, (body) => { + object(body) + check(typeof body.windowFrom === "number" && typeof body.windowTo === "number", "metrics should include its window") + check(typeof body.dlqEventsTotal === "number", "metrics should include the DLQ total") + }), + http.protected + .get("/oversight/trace", "oversight.trace") + .at((ctx) => ({ path: "/oversight/trace?correlationID=correlation_httpapi_missing", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + array(body.nodes) + }), + http.protected.get("/oversight/approvals", "oversight.approvals").json(200, (body) => { + object(body) + array(body.items) + }), + http.protected + .post("/oversight/approvals/resolve", "oversight.approvals.resolve") + .at((ctx) => ({ + path: "/oversight/approvals/resolve", + headers: ctx.headers(), + body: { id: "approval_httpapi_missing", decision: "acknowledged" }, + })) + .status(404, undefined, "status"), + http.protected + .post("/oversight/takeover", "oversight.takeover") + .mutating() + .at((ctx) => ({ + path: "/oversight/takeover", + headers: ctx.headers(), + body: { agentID: "agent_httpapi", reason: "HTTP API exercise" }, + })) + .json(200, (body) => { + object(body) + check( + typeof body.id === "string" && typeof body.workspaceID === "string", + "takeover should return its audit identity", + ) + check( + body.agentID === "agent_httpapi" && body.reason === "HTTP API exercise", + "takeover should preserve audit context", + ) + }), + http.protected + .post("/oversight/rollback", "oversight.rollback") + .at((ctx) => ({ + path: "/oversight/rollback", + headers: ctx.headers(), + body: { sessionID: "session_httpapi_missing", reason: "HTTP API exercise" }, + })) + .status(404, undefined, "status"), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/platform.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/platform.ts new file mode 100644 index 00000000..be674e88 --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/platform.ts @@ -0,0 +1,161 @@ +import { Effect } from "effect" +import { array, check, isRecord, object } from "../assertions" +import { http, route } from "../dsl" +import type { Scenario } from "../types" + +export const platformScenarios: Scenario[] = [ + http.protected + .get("/global/capabilities", "global.capabilities") + .global() + .json(200, (body) => { + object(body) + check(typeof body.protocolVersion === "string", "capabilities should include the protocol version") + check(typeof body.version === "string", "capabilities should include the build version") + object(body.features) + }), + http.protected.get("/global/projects", "global.projects").global().json(200, array), + http.protected + .delete("/global/projects/{projectID}", "global.projectDelete") + .global() + .mutating() + .at(() => ({ path: route("/global/projects/{projectID}", { projectID: "project_httpapi_missing" }) })) + .status(204, undefined, "status"), + http.protected + .post("/global/import", "global.import") + .global() + .probe({ path: "/global/import", body: { source: "invalid" } }) + .at(() => ({ path: "/global/import", body: { source: "invalid" } })) + .status(400), + http.protected + .post("/experimental/worktree/changes", "worktree.changes") + .seeded((ctx) => + Effect.acquireRelease(ctx.worktree({ name: "api-changes" }), (item) => ctx.worktreeRemove(item.directory)), + ) + .at((ctx) => ({ + path: "/experimental/worktree/changes", + headers: ctx.headers(), + body: { directory: ctx.state.directory }, + })) + .json(200, (body) => { + object(body) + check(typeof body.clean === "boolean", "worktree change count should include the clean decision") + check( + body.uncommitted === null || typeof body.uncommitted === "number", + "uncommitted count should be known or null", + ) + check(body.ahead === null || typeof body.ahead === "number", "ahead count should be known or null") + }), + http.protected + .post("/experimental/worktree/diff", "worktree.diff") + .seeded((ctx) => + Effect.acquireRelease(ctx.worktree({ name: "api-diff" }), (item) => ctx.worktreeRemove(item.directory)), + ) + .at((ctx) => ({ + path: "/experimental/worktree/diff", + headers: ctx.headers(), + body: { directory: ctx.state.directory }, + })) + .json(200, (body) => { + object(body) + array(body.entries) + check(typeof body.patch === "string", "worktree diff should include a patch") + check(typeof body.truncated === "boolean", "worktree diff should report truncation") + }), + http.protected + .post("/experimental/worktree/summary", "worktree.summary") + .seeded((ctx) => + Effect.acquireRelease(ctx.worktree({ name: "api-summary" }), (item) => ctx.worktreeRemove(item.directory)), + ) + .at((ctx) => ({ + path: "/experimental/worktree/summary", + headers: ctx.headers(), + body: { directory: ctx.state.directory }, + })) + .json(200, (body) => { + object(body) + check(typeof body.base === "string", "worktree summary should include its base branch") + check(typeof body.files === "number", "worktree summary should include the changed file count") + }), + http.protected + .post("/experimental/worktree/merge", "worktree.merge") + .at((ctx) => ({ + path: "/experimental/worktree/merge", + headers: ctx.headers(), + body: { directory: `${ctx.directory}/missing-worktree` }, + })) + .status(400), + http.protected + .delete("/experimental/worktree/safe-remove", "worktree.safeRemove") + .mutating() + .seeded((ctx) => + Effect.acquireRelease(ctx.worktree({ name: "api-safe-remove" }), (item) => ctx.worktreeRemove(item.directory)), + ) + .at((ctx) => ({ + path: "/experimental/worktree/safe-remove", + headers: ctx.headers(), + body: { directory: ctx.state.directory, force: true }, + })) + .json(200, (body) => check(body === true, "safe-remove should confirm removal")), + http.protected + .post("/session/{sessionID}/prompt_prepare", "session.prompt_prepare") + .seeded((ctx) => ctx.session({ title: "Prompt prepare validation" })) + .at((ctx) => ({ + path: route("/session/{sessionID}/prompt_prepare", { sessionID: ctx.state.id }), + headers: ctx.headers(), + body: { mode: "intelligence", parts: [{ type: "text", text: "" }] }, + })) + .status(400), + http.protected + .post("/session/{sessionID}/prompt_prepare_stream", "session.prompt_prepare_stream") + .stream() + .seeded((ctx) => ctx.session({ title: "Prompt prepare stream validation" })) + .at((ctx) => ({ + path: route("/session/{sessionID}/prompt_prepare_stream", { sessionID: ctx.state.id }), + headers: ctx.headers(), + body: { mode: "intelligence", parts: [{ type: "text", text: "" }] }, + })) + .status(200, (_ctx, result) => + Effect.sync(() => { + check(result.contentType.includes("text/event-stream"), "prompt prepare stream should use SSE") + check(result.text.includes("error"), "invalid prompt preparation should emit a terminal error event") + }), + ), + http.protected + .get("/session/{sessionID}/prompt_suggestion", "session.prompt_suggestion") + .seeded((ctx) => ctx.session({ title: "Prompt suggestion" })) + .at((ctx) => ({ + path: route("/session/{sessionID}/prompt_suggestion", { sessionID: ctx.state.id }), + headers: ctx.headers(), + })) + .json(200, (body) => { + object(body) + check(body.status === null, "new sessions should not have a prompt suggestion status") + check(body.body === null, "new sessions should not have a prompt suggestion body") + }), + http.protected + .get("/workspace/{workspaceID}/config/trusted-sources", "workspaceConfig.trustedSources.get") + .at((ctx) => ({ + path: route("/workspace/{workspaceID}/config/trusted-sources", { workspaceID: "workspace-httpapi" }), + headers: ctx.headers(), + })) + .json(200, (body) => { + object(body) + array(body.trustedSources) + check(body.trustedSources.includes("im"), "default trusted sources should include IM") + }), + http.protected + .put("/workspace/{workspaceID}/config/trusted-sources", "workspaceConfig.trustedSources.put") + .mutating() + .at((ctx) => ({ + path: route("/workspace/{workspaceID}/config/trusted-sources", { workspaceID: "workspace-httpapi" }), + headers: ctx.headers(), + body: { trustedSources: ["git", "ci"] }, + })) + .json(200, (body) => { + object(body) + check( + Array.isArray(body.trustedSources) && body.trustedSources.join(",") === "git,ci", + "trusted source replacement should round-trip", + ) + }), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/scenarios/runtime.ts b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/runtime.ts new file mode 100644 index 00000000..5094d7e3 --- /dev/null +++ b/packages/deepagent-code/test/server/httpapi-exercise/scenarios/runtime.ts @@ -0,0 +1,112 @@ +import { Effect } from "effect" +import { array, check, object } from "../assertions" +import { http } from "../dsl" +import type { Scenario } from "../types" + +export const runtimeScenarios: Scenario[] = [ + http.protected + .post("/debug/start", "debug.start") + .at((ctx) => ({ + path: "/debug/start", + headers: ctx.headers(), + body: { adapter: "httpapi-missing", program: "program.py" }, + })) + .json(200, (body) => { + object(body) + check(body.error === "adapter_unavailable", "unknown debug adapters should fail closed") + }), + http.protected + .post("/debug/breakpoints", "debug.breakpoints") + .at((ctx) => ({ + path: "/debug/breakpoints", + headers: ctx.headers(), + body: { sessionId: "debug_httpapi_missing", file: "main.ts", breakpoints: [{ line: 1 }] }, + })) + .status(500, undefined, "status"), + http.protected + .post("/debug/continue", "debug.continue") + .at((ctx) => ({ + path: "/debug/continue", + headers: ctx.headers(), + body: { sessionId: "debug_httpapi_missing" }, + })) + .status(500, undefined, "status"), + http.protected + .post("/debug/step", "debug.step") + .at((ctx) => ({ + path: "/debug/step", + headers: ctx.headers(), + body: { sessionId: "debug_httpapi_missing", kind: "next" }, + })) + .status(500, undefined, "status"), + http.protected + .get("/debug/stack", "debug.stack") + .at((ctx) => ({ path: "/debug/stack?sessionId=debug_httpapi_missing", headers: ctx.headers() })) + .status(500, undefined, "status"), + http.protected + .get("/debug/scopes", "debug.scopes") + .at((ctx) => ({ + path: "/debug/scopes?sessionId=debug_httpapi_missing&frameId=1", + headers: ctx.headers(), + })) + .status(500, undefined, "status"), + http.protected + .get("/debug/variables", "debug.variables") + .at((ctx) => ({ + path: "/debug/variables?sessionId=debug_httpapi_missing&variablesReference=1", + headers: ctx.headers(), + })) + .status(500, undefined, "status"), + http.protected + .post("/debug/evaluate", "debug.evaluate") + .at((ctx) => ({ + path: "/debug/evaluate", + headers: ctx.headers(), + body: { sessionId: "debug_httpapi_missing", expression: "value" }, + })) + .status(500, undefined, "status"), + http.protected + .post("/debug/terminate", "debug.terminate") + .at((ctx) => ({ + path: "/debug/terminate", + headers: ctx.headers(), + body: { sessionId: "debug_httpapi_missing" }, + })) + .status(500, undefined, "status"), + http.protected.get("/debug/sessions", "debug.sessions").json(200, (body) => { + object(body) + array(body.sessions) + }), + http.protected + .get("/debug/events", "debug.events") + .headersOnly() + .status(200, (_ctx, result) => + Effect.sync(() => { + check(result.contentType.includes("text/event-stream"), "debug events should use SSE") + }), + ), + http.protected + .post("/profile/run", "profile.run") + .at((ctx) => ({ + path: "/profile/run", + headers: ctx.headers(), + body: { program: "program.py", profiler: "httpapi-missing" }, + })) + .json(200, (body) => { + object(body) + check(body.status === "error", "an unavailable profiler should return a stable error result") + check(typeof body.runId === "string", "profile run should return a correlation id") + }), + http.protected + .get("/profile/result", "profile.result") + .at((ctx) => ({ path: "/profile/result?runId=profile_httpapi_missing", headers: ctx.headers() })) + .json(200, (body) => { + object(body) + check(body.status === "error" && body.error === "runId not found", "missing profile results should be explicit") + }), + http.protected + .get("/profile/hotspots", "profile.hotspots") + .at((ctx) => ({ path: "/profile/hotspots?runId=profile_httpapi_missing", headers: ctx.headers() })) + .json(200, array), + http.protected.get("/profile/runs", "profile.runs").json(200, array), +] diff --git a/packages/deepagent-code/test/server/httpapi-exercise/types.ts b/packages/deepagent-code/test/server/httpapi-exercise/types.ts index 2b94c2f3..5d0765ae 100644 --- a/packages/deepagent-code/test/server/httpapi-exercise/types.ts +++ b/packages/deepagent-code/test/server/httpapi-exercise/types.ts @@ -14,7 +14,7 @@ export type Method = (typeof Methods)[number] export type OpenApiMethod = (typeof OpenApiMethods)[number] export type Mode = "effect" | "coverage" | "auth" export type Comparison = "none" | "status" | "json" -export type CaptureMode = "full" | "stream" +export type CaptureMode = "full" | "headers" | "stream" export type AuthPolicy = "protected" | "public" | "public-bypass" | "ticket-bypass" export type ProjectOptions = { git?: boolean; config?: Partial; llm?: boolean } export type OpenApiSpec = { paths?: Record>> }