Skip to content

fix(skills): refresh snapshots after skill exposure config changes#82365

Closed
DevInvestor wants to merge 1 commit into
openclaw:mainfrom
DevInvestor:dan/xpost-snapshot-runtime-20260515
Closed

fix(skills): refresh snapshots after skill exposure config changes#82365
DevInvestor wants to merge 1 commit into
openclaw:mainfrom
DevInvestor:dan/xpost-snapshot-runtime-20260515

Conversation

@DevInvestor

@DevInvestor DevInvestor commented May 16, 2026 •
edited
Loading

Copy link
Copy Markdown

Summary

  • add a redacted skill snapshot config fingerprint so persisted skillsSnapshot entries refresh after skill exposure config changes, including skills.load.extraDirs
  • persist the fingerprint on built skill snapshots and compare it in auto-reply/session update and agent-command paths
  • add regression coverage for Tom/HQ-style shared skill exposure where xurl is local but x-post-analysis appears only via shared config

Context

This is Lane A for GB-P12-W4 x-post-analysis deterministic workflow. It fixes the runtime side of the incident where Tom's persistent session had xurl but not x-post-analysis, so it could not load the built X-post workflow.

Real behavior proof

  • Behavior or issue addressed: persisted agent skillsSnapshot entries now refresh when the runtime skill exposure config changes, including skills.load.extraDirs, instead of keeping a stale snapshot that has xurl but not x-post-analysis.
  • Real environment tested: Clawdbot-1 Ubuntu OpenClaw checkout at /root/clawd/.worktrees/dan-xpost-snapshot-runtime, Node v22.22.2, using real local skill directories and /root/clawd/skills/x-post-analysis/SKILL.md as the shared skill exposed through skills.load.extraDirs.
  • Exact steps or command run after this patch: 
    cd /root/clawd/.worktrees/dan-xpost-snapshot-runtime
node --import tsx ./.tmp-xpost-real-behavior-proof.ts
  • Evidence after fix: terminal output from the real OpenClaw checkout: 
    {
  "setup": {
    "workspace": "/tmp/openclaw-xpost-proof-eaZ0Io/workspace",
    "managedSkillsDir": "/tmp/openclaw-xpost-proof-eaZ0Io/managed-skills",
    "bundledSkillsDir": "/tmp/openclaw-xpost-proof-eaZ0Io/bundled-skills-empty",
    "sharedSkillsExtraDir": "/root/clawd/skills",
    "agentId": "tom",
    "skillFilter": ["xurl", "x-post-analysis"]
  },
  "before": {
    "extraDirs": [],
    "skills": ["browser-automation", "xurl"],
    "hasXurl": true,
    "hasXPost": false,
    "configFingerprint": "de6316be2f21..."
  },
  "after": {
    "extraDirs": ["/root/clawd/skills"],
    "skills": ["brave-search", "browser-automation", "coding-agent", "elevenlabs-voices", "github", "notion", "phone-agent", "planning-with-files", "video-transcript-downloader", "weather", "x-post-analysis", "xurl"],
    "hasXurl": true,
    "hasXPost": true,
    "xpostResolved": "/root/clawd/skills/x-post-analysis/SKILL.md",
    "configFingerprint": "3f0f7ab04b1c..."
  },
  "fingerprintChanged": true
}
  • Observed result after fix: after adding /root/clawd/skills to skills.load.extraDirs, the real snapshot builder exposes x-post-analysis and records a changed redacted configFingerprint, which is the persisted signal used by the patch to refresh stale skillsSnapshot data.
  • What was not tested: no live gateway restart, no live Tom session mutation, no production config edit, and no deploy in this PR.

Verification

  • OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/test-projects.mjs --maxWorkers=1 src/agents/skills.buildworkspaceskillsnapshot.test.ts
  • OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/test-projects.mjs --maxWorkers=1 src/auto-reply/reply/session-updates.test.ts
  • OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/test-projects.mjs --maxWorkers=1 src/agents/agent-command.live-model-switch.test.ts
  • node scripts/run-oxlint.mjs --tsconfig config/tsconfig/oxlint.core.json ...
  • node scripts/run-tsgo.mjs -p test/tsconfig/tsconfig.test.src.json ...
  • Riley independent review: APPROVED

Deployment

No live runtime/config/session mutation in this PR. Deployment/config exposure remains separate and must include backups, health checks, Tom hasXPost=true proof, and rollback.

@openclaw-barnacle openclaw-barnacle Bot added agents Agent runtime and tooling size: M triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 16, 2026
@clawsweeper

clawsweeper Bot commented May 16, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Codex review: needs maintainer review before merge.

Summary
The PR adds a redacted skill snapshot config fingerprint, persists it on skill snapshots, compares it in auto-reply and agent-command paths, and adds regression coverage for skills.load.extraDirs exposure changes.

Reproducibility: yes. At source level, current main builds snapshots from skills.load.extraDirs but persisted snapshots are reused unless version or filter state changes, so a config-only exposure change can leave a stale snapshot.

Real behavior proof
Sufficient (terminal): The PR body now includes terminal output from a real Ubuntu checkout showing the before/after skill exposure and changed fingerprint after adding skills.load.extraDirs.

Next step before merge
No repair lane is needed; the PR appears focused and should proceed through ordinary maintainer and CI review.

Security
Cleared: No concrete security or supply-chain concern found; the diff adds no dependencies or workflows and persists only a SHA-256 fingerprint of redacted config.

Review details

Best possible solution:

Land the narrow fingerprint-based invalidation after normal CI and maintainer checks, keeping live deployment and Tom/HQ config exposure as a separate rollout step.

Do we have a high-confidence way to reproduce the issue?

Yes. At source level, current main builds snapshots from skills.load.extraDirs but persisted snapshots are reused unless version or filter state changes, so a config-only exposure change can leave a stale snapshot.

Is this the best way to solve the issue?

Yes. Persisting and comparing a redacted config fingerprint in both snapshot reuse paths is the narrow maintainable fix, and the helper keeps raw config secrets out of the persisted field.

What I checked:

  • Current main refresh condition: Current main refreshes persisted auto-reply skill snapshots for version or skill-filter changes only, while the config fingerprint is only used in the in-memory resolved-skills cache key. (src/auto-reply/reply/session-updates.ts:264, efabae2f9bf0)
  • Current main extraDirs source: The snapshot builder already reads skills.load.extraDirs, so config changes can alter the correct snapshot content without changing the snapshot version or filter. (src/agents/skills/workspace.ts:848, efabae2f9bf0)
  • Current main agent-command condition: The agent-command path also reuses persisted snapshots unless the snapshot is missing, stale by version, or has a different skill filter. (src/agents/agent-command.ts:649, efabae2f9bf0)
  • PR implementation: The PR adds snapshot-fingerprint.ts, stores configFingerprint on snapshots, and compares it before reusing persisted snapshots in both auto-reply and agent-command paths. (src/agents/skills/snapshot-fingerprint.ts:59, 67998033e3ef)
  • Regression coverage: The diff adds tests for shared x-post-analysis discovery through skills.load.extraDirs and for refreshing a stale persisted xurl-only snapshot after config-driven exposure changes. (src/auto-reply/reply/session-updates.test.ts:239, 67998033e3ef)
  • Real behavior proof: The updated PR body includes copied terminal JSON from a real Ubuntu checkout showing x-post-analysis absent before adding /root/clawd/skills, present after, and a changed redacted config fingerprint; GitHub's Real behavior proof check then completed successfully. (67998033e3ef)

Likely related people:

  • steipete: Related PR metadata and local blame tie the recent agent/session runtime and snapshot-cache behavior that this PR extends to steipete/Peter Steinberger. (role: recent area contributor and merger; confidence: high; commits: 8ba0bb2a8a9b, d643cd5a9f08, 628c753f3bff; files: src/auto-reply/reply/session-updates.ts, src/agents/skills/workspace.ts, src/agents/agent-command.ts)
  • Josh Avant: Current local history shows recent work on src/agents/agent-command.ts, one of the paths this PR updates, though not the central skill snapshot logic. (role: recent adjacent contributor; confidence: low; commits: 1dac68c0bbe7, 64b94daf92c4; files: src/agents/agent-command.ts)

Codex review notes: model gpt-5.5, reasoning high; reviewed against efabae2f9bf0.

@openclaw-barnacle openclaw-barnacle Bot added proof: supplied External PR includes structured after-fix real behavior proof. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 16, 2026
@clawsweeper clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 16, 2026
This was referenced May 19, 2026
Merged
Closed
@DevInvestor DevInvestor closed this Jun 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

agents Agent runtime and tooling proof: sufficient ClawSweeper judged the real behavior proof convincing. proof: supplied External PR includes structured after-fix real behavior proof. size: M

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@DevInvestor