Repository navigation
Conversation
Layer (b) of the authoring platform (iflytek#934), stacking on iflytek#935: - **Runtime bindings**: agent type + config + tool allowlist + MCP server declarations per draft, persisted as JSONB and validated at save time; local-script and OpenAI-compatible agent types - **CONFIG and BEHAVIOR validation layers**: validation.yaml spec parsing (defensive YAML), per-task execution through the runtime adapters (inline subprocesses with a cleared environment, or the docker backend with a locked-down container), MCP servers probed for real during validation with tool discovery surfaced in the event stream; assertion evaluation (exit codes, stdout/stderr contains, timeouts) - **Security policies** (the layer that wants the dedicated review): the shared AuthoringSecurityPolicy gates OpenAI-compatible and http/sse MCP endpoints at save time and again at connect time — cloud-metadata and link-local addresses are always rejected, loopback/RFC1918/ULA/ CGNAT/RFC2544 ranges sit behind per-surface flags; stdio MCP transport is disabled by default and, in docker execution mode, wraps its command in a hardened container (no network, dropped capabilities, read-only root, resource caps) that is torn down on close; envRefs may only name variables in the operator-configured allowlist; docker execution-mode is the default and inline mode fails startup outside local/dev/test; HTTP redirects are never followed and response/stdio line sizes are capped at 2 MiB. Verified: security-policy and MCP unit tests (policy classification per surface, redirect refusal, response caps, envRef filtering, docker-wrapped commands), adapter tests including a real-container isolation check, and the runtime flow integration test (three-layer validation with inline script execution; cloud-metadata/envRefs/LLM-surface bindings rejected at save time). Signed-off-by: zjncs <18910855655@163.com>
…ime orchestrator Rebases the runtime layer onto main, which now carries iflytek#935 plus the follow-up hardening (FutureTask per run with duplicate-submit fencing, a wall-clock watchdog on the timer, settle-first cancellation with worker interrupt, appendTerminalEvent, revision-pinned materializeSnapshot, and the PackageEntry streaming API). The three-phase orchestrator keeps all of it: the watchdog remains the hard deadline while the behavior layer keeps its between-task soft deadline; the cooperative cancel signal is now the worker's interrupt status; the config/behavior layers read the pinned snapshot and the draft id from the run instead of a reloaded draft. The upstream orchestrator test is adapted to the runtime-layer constructor (no adapters registered — the lifecycle paths under test never reach the config layer) and passes unchanged in its assertions. Signed-off-by: zjncs <18910855655@163.com>
…s deletable The validation container drops every capability, including DAC_OVERRIDE, so the mounted workspace must be world-readable on the host for the in-container interpreter to open the script, and directories must be traversable and writable so the script can create artifacts. Fresh JVM temp directories are owner-only on Linux (umask 077 runners, hardened hosts), which broke script execution with 'sh: can't open: Permission denied' — macOS Docker Desktop masks this by remapping ownership inside its VM, which is why the local suite passed while the hosted-runner CI failed. The container also runs as the host JVM's uid:gid (--user) so files the script creates inside the mount belong to the same user that cleans the workspace up; running as the image-default root left artifacts that the host user (and the test harness) could not delete. Signed-off-by: zjncs <18910855655@163.com>
The push of this rebase did not create a workflow run (GitHub-side hiccup); an empty commit to retrigger the backend-tests check. Signed-off-by: zjncs <18910855655@163.com>
Picks up upstream API drift (S3 storage options, github-sync, dsh 0.3.1) on top of the authoring endpoints. Signed-off-by: zjncs <18910855655@163.com>
Layer (c) of the authoring platform (iflytek#934), stacking on the runtime layer: - **Fix loop**: apply a finding's suggested patch to the draft in one call (sha256-guarded old content, revision bump, validation verdict invalidated by the new revision) or dismiss it as not applicable - **Submission**: validated drafts submit into the existing SkillHub publish pipeline — the gate is strict (the draft's current revision must be bound to a run that succeeded with zero errors), so content can never be published without the exact bytes that were validated; reuses SkillPublishService (package re-validation, security scan, review flow) and records the resulting skill id on the draft Verified by the full authoring flow integration test: create → edit → bind runtime → three-layer validation → fix a finding → re-validate → submit into the (mocked) publish pipeline. Signed-off-by: zjncs <18910855655@163.com>
Adds the finding apply/dismiss and draft submit endpoints on top of the runtime-layer schema. Signed-off-by: zjncs <18910855655@163.com>
Layer (d) of the authoring platform (iflytek#934), stacking on the fix/publish layer: the web workbench under /dashboard/authoring — - draft list and draft detail (files / runtime binding / validation runs) with an editor that tracks unsaved state and surfaces optimistic- concurrency conflicts - runtime binding form switching fields by agent type (interpreter vs endpoint/model), tool allowlist and MCP server declarations with immediate validation - run detail: status card, live event console (SSE with a 2s polling fallback and cursor replay), finding cards with severity, location, suggested-fix diff preview and one-click apply/dismiss - submit dialog handing validated drafts to the publish pipeline - copy in en/zh/ru; Playwright e2e for the full authoring flow (e2e/authoring-flow.spec.ts) Verified: typecheck clean; 922 web unit tests pass (the only failures in the suite are the two pre-existing upstream theme tests, which fail identically on plain main). Signed-off-by: zjncs <18910855655@163.com>
…scripts The platform's design and operations doc — domain model and tables, REST API surface, frontend workbench, deployment configuration and security baseline — plus the recorded evidence: the end-to-end publish run and the real-board RISC-V64 verification (the board workflow runs on the contributor fork; the doc links the green runs and the setup scripts ship with the repo for re-verification). Also ships the example skill (SKILL.md, script, validation.yaml, references), the live verification scripts (smoke 39 checks, publish E2E 16 checks) and the RISC-V64 re-verification scripts. The migration is tracked as V69 with its renumbering history kept for context. Signed-off-by: zjncs <18910855655@163.com>
There was a problem hiding this comment.
Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.
An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.
Contributor
Author
|
Closing in favor of the layer-by-layer cadence: stacked-against-main PRs show the cumulative diff (this one included everything below it), which defeats the point of the split. This layer will be resubmitted against main as soon as the layer below merges, at which point the diff collapses to this layer's own changes. The incremental diff stays reviewable today at the fork-stacked PR linked from #934. Sorry for the churn. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Layer (e) — the final layer — of the authoring platform split (#934), stacking on #954 (b), #1000 (c) and the web layer (d).
What this layer adds
docs/26-skill-authoring-platform.md: the platform's design and operations doc — domain model and tables, REST API surface, frontend workbench, deployment configuration (allskillhub.authoring.*keys), and the security baseline (SSRF posture, stdio MCP isolation, envRefs allowlist, docker default + startup guard). Includes the recorded evidence: the end-to-end publish run and the real-board RISC-V64 verification — the board workflow runs on the contributor fork (no board runners upstream), the doc links the green runs as the architecture evidence, and the setup scripts ship in-repo for re-verification.examples/skill-authoring/: a complete example skill (SKILL.md, executable script, validation.yaml, references) that doubles as a walkthrough of the authoring flow.authoring-smoke-test.sh(39 checks over HTTP against a running server) andauthoring-publish-e2e.sh(16 checks: create → validate → submit → scan → review → PUBLISHED).riscv64-verify.sh(QEMU/native auto-detect, docker or jar boot modes) andriscv64-native-setup.sh(one-shot provisioning on a real riscv64 board).With this layer the stack is complete: (a) domain + CRUD + structure validation → (b) runtime + MCP + security → (c) fix loop + publish → (d) web workbench → (e) docs, example and scripts.