dirge-code
diff --git a/‎.beads/issues.jsonl‎
Lines changed: 1 addition & 1 deletion b/‎.beads/issues.jsonl‎
Lines changed: 1 addition & 1 deletion
diff --git a/‎src/ui/renderer.rs‎
Lines changed: 189 additions & 0 deletions b/‎src/ui/renderer.rs‎
Lines changed: 189 additions & 0 deletions
@@ -25,7 +25,7 @@
 {"_type":"issue","id":"dirge-86e","title":"ANSI injection in permission ALERT prompt","description":"ask_req.tool / ask_req.input rendered un-sanitized at mod.rs:2584-2585. Reopen path already sanitizes — asymmetric. Sec impl: ANSI at the permission-decision moment.","status":"closed","priority":1,"issue_type":"bug","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-21T22:17:34Z","created_by":"Yogthos","updated_at":"2026-05-21T22:26:37Z","started_at":"2026-05-21T22:17:42Z","closed_at":"2026-05-21T22:26:37Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-9f1","title":"Chat history ignores 120-col content_width cap","description":"max_line_width and wrap_line use raw content_cols, so on wide terminals scrollback overflows the centered band into divider/panel margin.","status":"closed","priority":1,"issue_type":"bug","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-21T22:17:33Z","created_by":"Yogthos","updated_at":"2026-05-21T22:26:36Z","started_at":"2026-05-21T22:17:42Z","closed_at":"2026-05-21T22:26:36Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-woq","title":"R1: fix 3 critical plugin bugs (FFI panic, dialog deadlock, init hang)","description":"From the plugin subsystem audit: (1) wrap JanetCFunctions in catch_unwind so Rust panics don't unwind across the C-FFI boundary into Janet; (2) cancel send_dialog's reply_rx.recv() on worker shutdown so the worker thread doesn't block forever when the UI exits mid-dialog; (3) add timeout to the init handshake so a worker panic before init_tx.send() doesn't hang the main thread. Also: (4) bounds-assert wrap_string's i32 cast for the unlikely \u003e2GB case, (5) make take_string_slot atomic to close the race window, (6) don't eat unrelated user events in the dialog arm.","status":"closed","priority":1,"issue_type":"bug","owner":"yogthos@gmail.com","created_at":"2026-05-20T14:59:57Z","created_by":"Yogthos","updated_at":"2026-05-20T15:30:28Z","started_at":"2026-05-20T15:00:10Z","closed_at":"2026-05-20T15:30:28Z","dependency_count":0,"dependent_count":1,"comment_count":0}
-{"_type":"issue","id":"dirge-ov2","title":"UI: subagent multi-chat-window (Ctrl-N/P, /tasks picker) — port from maki","description":"Each subagent task gets its own chat window; navigate via Ctrl-N (next) / Ctrl-P (previous) and /tasks slash command opens a picker. Maki: maki-ui/src/app/mod.rs:317-322 (open_tasks), :341-352 (Ctrl-N/P), components/keybindings.rs:135-136 (NEXT_CHAT/PREV_CHAT), components/command.rs:22-23 (/tasks).\n\nBigger refactor: dirge's renderer holds ONE buffer. Need to model Vec\u003cChat\u003e with active_chat index, swap renderer buffer on switch. Conflict: dirge's Ctrl-X currently drops queued interjections — would need a different binding for next-task, or rebind interjection-drop. Coordinate with user before starting.","status":"open","priority":2,"issue_type":"feature","owner":"yogthos@gmail.com","created_at":"2026-05-23T17:37:07Z","created_by":"Yogthos","updated_at":"2026-05-23T17:37:07Z","dependency_count":0,"dependent_count":0,"comment_count":0}
+{"_type":"issue","id":"dirge-ov2","title":"UI: subagent multi-chat-window (Ctrl-N/P, /tasks picker) — port from maki","description":"Each subagent task gets its own chat window; navigate via Ctrl-N (next) / Ctrl-P (previous) and /tasks slash command opens a picker. Maki: maki-ui/src/app/mod.rs:317-322 (open_tasks), :341-352 (Ctrl-N/P), components/keybindings.rs:135-136 (NEXT_CHAT/PREV_CHAT), components/command.rs:22-23 (/tasks).\n\nBigger refactor: dirge's renderer holds ONE buffer. Need to model Vec\u003cChat\u003e with active_chat index, swap renderer buffer on switch. Conflict: dirge's Ctrl-X currently drops queued interjections — would need a different binding for next-task, or rebind interjection-drop. Coordinate with user before starting.","notes":"## Design (post-scope-decision: user picked Heavier refactor)\n\nFive phases, each independently committable so we can ship partial\nprogress if needed.\n\n### Phase A: Multi-buffer in Renderer (~200-300 lines)\n\nPer-chat state extracted into a `ChatBuffer` struct:\n- buffer: Vec\u003cLineEntry\u003e\n- partial: CompactString + partial_color\n- scroll_offset, lines, col\n- selection_active, selection_start, selection_end\n\nGlobal state stays on Renderer:\n- input_rows, monochrome\n- panel_mode, panel_data\n- avatar_state, avatar_tick\n\nRenderer gains:\n- chats: Vec\u003cChatBuffer\u003e\n- active_chat: usize\n- add_chat(name) -\u003e ChatId\n- switch_chat(id)\n- chat_names()\n\nAll existing public methods (write_line, write, replace_from,\nrender_viewport, scroll_*, selection_*) route to chats[active_chat].\n\nBackward compat: Renderer::new() creates one default chat (active=0).\nSingle-chat usage = existing behavior.\n\n### Phase B: Keybinds + /tasks (~80 lines)\n\n- Ctrl-N: next chat\n- Ctrl-P: prev chat\n- Ctrl-X: open /tasks picker (rebind interjection-drop to Alt+X\n  per ov2 plan)\n- `/tasks` slash command: same picker\n\nPicker shows chat names; selection switches active.\n\n### Phase C: Per-chat UI state (~150 lines)\n\nui/mod.rs has 10+ state variables that are currently per-session:\n- response_buf, response_start_line\n- reasoning_buf, reasoning_start_line\n- last_tool_name, last_tool_call_id\n- tool_chamber_open, agent_line_started\n- was_reasoning, tool_calls_buf\n\nMove to a `ChatUiState` struct, one per chat. The UI loop reads\nstate from `chat_states[active_chat]` on each event.\n\n### Phase D: task tool refactor (~200 lines)\n\nCurrently `task` tool calls `model.btw_query(prompt)` — one-shot,\nno tools. Refactor to:\n- Spawn a sub-runner using same provider but limited tool set\n  (read-only? or same as parent? — needs decision)\n- Subagent emits AgentEvents to its own mpsc channel\n- task tool returns the chat_id; UI routes events from the channel\n  to chats[chat_id]\n\n### Phase E: Event router (~80 lines)\n\nUI loop selects across multiple agent_rx channels (one per active\nsubagent). Each event lands in chat_states[event.chat_id] and\nrenders via chats[event.chat_id].\n\n## Open question for Phase D\n\nShould subagent inherit parent's tool set, or be restricted to\nread-only? Maki gives subagents full tools by default. Dirge's\npermission system already gates per-tool, so safety isn't broken\neither way. Inheriting is simpler.","status":"in_progress","priority":2,"issue_type":"feature","owner":"yogthos@gmail.com","created_at":"2026-05-23T17:37:07Z","created_by":"Yogthos","updated_at":"2026-05-23T17:59:28Z","started_at":"2026-05-23T17:53:50Z","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-rmk","title":"CLI: --output-format stream-json headless mode (port from maki)","description":"dirge has --print boolean but no output format selector. Port from maki:\n- maki/src/main.rs:32-58 (Cli with --print + --output-format)\n- maki/src/print.rs:44-49 (OutputFormat enum: Text | Json | StreamJson)\n- maki/src/print.rs:51-110 (PrintResult / InitEvent / AssistantEvent / UserEvent structs — Claude-Code-compatible JSON schema)\n- maki/src/print.rs:126-388 (headless run loop with NDJSON event serialization at lines 280, 285, 304, 325, 334)\n\nAdapt: dirge's run_print already exists at src/provider.rs:572. Extend it with format dispatch. For stream-json: serde-serialize each AgentEvent (or a JSON-shaped envelope) and println! one object per line.","status":"closed","priority":2,"issue_type":"feature","owner":"yogthos@gmail.com","created_at":"2026-05-23T17:37:06Z","created_by":"Yogthos","updated_at":"2026-05-23T17:51:48Z","started_at":"2026-05-23T17:44:18Z","closed_at":"2026-05-23T17:51:48Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-efw","title":"Perm review F3: space-star pattern optional-arg semantics","description":"LOW-polish gap found in opencode-vs-dirge review. Opencode's wildcard matcher (util/wildcard.ts:13-15) rewrites trailing ' *' as '( .*)?' — making the trailing args optional. So 'ls *' matches BOTH 'ls' and 'ls -la'. Dirge's pattern.rs glob_to_regex requires the args (' *' becomes ' .*' which needs at least the space).\n\nConcrete UX friction: a user accepts 'ls *' from a session allowlist after the agent runs 'ls -la', then the agent runs bare 'ls' — re-prompted because the saved pattern doesn't match. Port the optional-suffix rewrite to src/permission/pattern.rs's glob_to_regex.","status":"closed","priority":2,"issue_type":"task","owner":"yogthos@gmail.com","created_at":"2026-05-23T15:48:20Z","created_by":"Yogthos","updated_at":"2026-05-23T16:07:24Z","started_at":"2026-05-23T15:57:07Z","closed_at":"2026-05-23T16:07:24Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-84j","title":"B3-9: edit tool fuzzy-match cascade","description":"edit.rs:155-160 returns 'old_text not found' on whitespace/indent/CRLF drift. LLMs frequently hit this. opencode edit.ts:222-432 has simple → lineTrimmed → whitespace-normalized → indentation-flexible → levenshtein. pi edit-diff.ts:91-132 has fuzzyFindText. Port the cascade.","status":"closed","priority":2,"issue_type":"feature","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-23T03:24:44Z","created_by":"Yogthos","updated_at":"2026-05-23T03:43:39Z","started_at":"2026-05-23T03:37:16Z","closed_at":"2026-05-23T03:43:39Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 
@@ -58,6 +58,29 @@ pub struct PanelData {
     pub modified: Vec<String>,
 }
 
+/// Per-chat state saved while a chat is INACTIVE. Mirrors the fields
+/// the active chat uses on the `Renderer` itself; switching chats
+/// swaps state in/out via `save_active` / `load_active`. Keeps the
+/// hot-path rendering code unchanged — only chat-switch boundaries
+/// pay the snapshot cost.
+///
+/// dirge-ov2 Phase A: enables multiple subagent chat windows. The
+/// main session is always at index 0; subagent chats start at index
+/// 1. Selection state lives per-chat because a selection in chat A
+/// would be meaningless when chat B is on screen.
+pub struct ChatSnapshot {
+    pub name: String,
+    buffer: Vec<LineEntry>,
+    partial: CompactString,
+    partial_color: Color,
+    scroll_offset: usize,
+    lines: u16,
+    col: u16,
+    selection_active: bool,
+    selection_start: Option<(usize, usize)>,
+    selection_end: Option<(usize, usize)>,
+}
+
 pub struct Renderer {
     lines: u16,
     col: u16,
@@ -66,6 +89,13 @@ pub struct Renderer {
     partial: CompactString,
     partial_color: Color,
     scroll_offset: usize,
+    /// dirge-ov2: snapshots of the OTHER chats — the active chat's
+    /// state lives in the fields above. `chats[active_chat]` is the
+    /// "free slot" (its name/buffer match what's on screen but the
+    /// fields haven't been written into it yet; switching chats
+    /// flushes them).
+    chats: Vec<ChatSnapshot>,
+    active_chat: usize,
     /// Number of rows the input area currently occupies (1 by default, grows
     /// up to MAX_INPUT_VISIBLE_LINES as the user adds newlines or types past
     /// the wrap width). The chat viewport shrinks by the same amount.
@@ -100,6 +130,11 @@ impl Renderer {
             partial: CompactString::new(""),
             partial_color: Color::White,
             scroll_offset: 0,
+            // dirge-ov2: one default "main" chat. Subagent chats are
+            // appended via `add_chat`. Index 0 is always the main
+            // session.
+            chats: vec![ChatSnapshot::empty("main")],
+            active_chat: 0,
             input_rows: 1,
             monochrome: false,
             selection_active: false,
@@ -112,6 +147,103 @@ impl Renderer {
         })
     }
 
+    /// dirge-ov2: append a new chat (typically a subagent) with the
+    /// supplied display name. Returns the new chat's index, which the
+    /// caller stores so it can target events at this chat later via
+    /// `switch_chat`.
+    ///
+    /// The new chat starts empty — no buffer entries, no selection,
+    /// no scroll. Does NOT switch to it; the caller chooses when to
+    /// surface the new chat in the UI.
+    pub fn add_chat(&mut self, name: impl Into<String>) -> usize {
+        self.chats.push(ChatSnapshot::empty(name.into()));
+        self.chats.len() - 1
+    }
+
+    /// dirge-ov2: switch the active chat. Saves the current chat's
+    /// state to its snapshot, loads the target chat's snapshot into
+    /// the Renderer's hot fields, and triggers a viewport repaint via
+    /// the next render call. No-op if `idx == active_chat`.
+    pub fn switch_chat(&mut self, idx: usize) {
+        if idx == self.active_chat || idx >= self.chats.len() {
+            return;
+        }
+        self.save_active();
+        self.active_chat = idx;
+        self.load_active();
+    }
+
+    pub fn active_chat(&self) -> usize {
+        self.active_chat
+    }
+
+    pub fn chat_count(&self) -> usize {
+        self.chats.len()
+    }
+
+    pub fn chat_names(&self) -> Vec<String> {
+        // Active chat's name lives in `chats[active_chat]` too (kept
+        // in sync at add-time; mutations of the active chat's name
+        // would go through a dedicated setter if added later).
+        self.chats.iter().map(|c| c.name.clone()).collect()
+    }
+
+    /// dirge-ov2: snapshot the current hot fields into the active
+    /// chat's slot. Called before switching chats and when the
+    /// caller wants a consistent persistent state (e.g. session
+    /// save).
+    fn save_active(&mut self) {
+        let slot = &mut self.chats[self.active_chat];
+        slot.buffer = std::mem::take(&mut self.buffer);
+        slot.partial = std::mem::take(&mut self.partial);
+        slot.partial_color = self.partial_color;
+        slot.scroll_offset = self.scroll_offset;
+        slot.lines = self.lines;
+        slot.col = self.col;
+        slot.selection_active = self.selection_active;
+        slot.selection_start = self.selection_start;
+        slot.selection_end = self.selection_end;
+    }
+
+    /// dirge-ov2: load the active chat's snapshot into the hot
+    /// fields. Inverse of `save_active`. Called after `switch_chat`
+    /// updates `active_chat`.
+    fn load_active(&mut self) {
+        let slot = &mut self.chats[self.active_chat];
+        self.buffer = std::mem::take(&mut slot.buffer);
+        self.partial = std::mem::take(&mut slot.partial);
+        self.partial_color = slot.partial_color;
+        self.scroll_offset = slot.scroll_offset;
+        self.lines = slot.lines;
+        self.col = slot.col;
+        self.selection_active = slot.selection_active;
+        self.selection_start = slot.selection_start;
+        self.selection_end = slot.selection_end;
+    }
+}
+
+impl ChatSnapshot {
+    fn empty(name: impl Into<String>) -> Self {
+        Self {
+            name: name.into(),
+            buffer: Vec::new(),
+            partial: CompactString::new(""),
+            partial_color: Color::White,
+            scroll_offset: 0,
+            lines: 0,
+            col: 0,
+            selection_active: false,
+            selection_start: None,
+            selection_end: None,
+        }
+    }
+}
+
+#[allow(dead_code)]
+impl Renderer {
+    // (continuation marker — methods below this block remain unchanged)
+    fn _ov2_phase_a_anchor() {}
+
     /// Update the avatar state and trigger a repaint of the bottom-left
     /// pixels. Cheap when the state hasn't changed — only the existing
     /// 3-row × 5-col patch is re-drawn.
@@ -1635,6 +1767,63 @@ pub fn copy_to_clipboard(text: &str) {
 mod tests {
     use super::*;
 
+    /// dirge-ov2 Phase A: chat switching saves the prior chat's
+    /// buffer and selection, then loads the target chat's snapshot.
+    /// Round-trip preserves content.
+    #[test]
+    fn chat_snapshot_save_load_roundtrip() {
+        let mut r = Renderer::new().expect("renderer");
+        // Default chat is "main" at index 0.
+        assert_eq!(r.active_chat(), 0);
+        assert_eq!(r.chat_count(), 1);
+        assert_eq!(r.chat_names(), vec!["main".to_string()]);
+
+        // Seed main chat with some content.
+        r.buffer.push(LineEntry {
+            text: CompactString::new("main-line-1"),
+            color: Color::White,
+        });
+        r.scroll_offset = 5;
+
+        // Spawn a subagent chat and switch to it.
+        let sub_idx = r.add_chat("subagent-1");
+        assert_eq!(sub_idx, 1);
+        assert_eq!(r.chat_count(), 2);
+        r.switch_chat(sub_idx);
+        assert_eq!(r.active_chat(), 1);
+
+        // Subagent chat starts empty.
+        assert!(r.buffer.is_empty());
+        assert_eq!(r.scroll_offset, 0);
+
+        // Add content to the subagent chat.
+        r.buffer.push(LineEntry {
+            text: CompactString::new("sub-line-1"),
+            color: Color::Cyan,
+        });
+        r.scroll_offset = 2;
+
+        // Switch back to main — its content must be restored.
+        r.switch_chat(0);
+        assert_eq!(r.buffer.len(), 1);
+        assert_eq!(r.buffer[0].text.as_str(), "main-line-1");
+        assert_eq!(r.scroll_offset, 5);
+
+        // Switch back to subagent — its content also restored.
+        r.switch_chat(1);
+        assert_eq!(r.buffer.len(), 1);
+        assert_eq!(r.buffer[0].text.as_str(), "sub-line-1");
+        assert_eq!(r.scroll_offset, 2);
+
+        // Switch to same chat is a no-op.
+        r.switch_chat(1);
+        assert_eq!(r.buffer.len(), 1);
+
+        // Out-of-range index is a no-op (defensive — caller bug).
+        r.switch_chat(99);
+        assert_eq!(r.active_chat(), 1);
+    }
+
     /// Create a renderer with a synthetic buffer of `n` short lines so we
     /// can drive scroll/append behavior without touching a real terminal.
     fn fresh_with_lines(n: usize) -> Renderer {