feat(ce-skill-work): repo-local skill for authoring, editing, reviewing, and responding to review on skills - #1402
Conversation
…ing, and responding to review on skills Move the skill-lifecycle procedures out of always-loaded AGENTS.md into a repo-local skill (.claude/skills/skill-work, symlinked at .agents/skills/skill-work for Codex) that loads when skill work starts. AGENTS.md keeps the standard, the reviewer rules bots read, and a pointer. The skill adds provenance-gated removal, an audit-question table, the over-cut guard, trigger-only descriptions, per-step done criteria, and a validation contract. Not part of the distributed plugin. Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
PR SummaryCursor Bugbot is generating a summary for commit 5d87499. Configure here. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5d87499fcc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- provenance triggers investigation; admission decides retention (not an automatic keep/cut) - completion report shaped per mode; review mode reports findings and checked paths only - resolver rubric invokes a project-routed procedure for the file class before judging - guard pins the four-activity routing wording and each mode->reference mapping Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
…to ce-skill-work Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
tmchow
changed the title
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2d93559666
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
- evaluate: forward-test hygiene (agent must not know it is a test; artifacts not conclusions; fresh context; clean up), substantive activation prompts, read transcripts, generalize not overfit - new-skill: pushier activation contract for under-triggering, naming rules, match specificity to fragility, one-level references / TOC / no auxiliary files - edit-skill: shouting (ALWAYS/NEVER) as an audit smell - description names the phrasings that should trigger without the skill's name Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
- edit-skill step 2 defers to SKILL.md's sediment rule instead of restating a keep/cut verdict; rg for the search example Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ce68e2f6f9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- evaluate: invoke skill-creator through the harness's skill mechanism, not a slash form - review: Verify/Consider are advisory; findings-only transports get Change items only Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
- edit-skill step 2 defers to the sediment rule without paraphrasing it Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1ee637bc2f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…where they fire - edit/review: decentering step (state runtime, distrust "redundant" and "missing X"), protocol-vs-judgment disappearance test with the coverage-menu guard, deviation record, scope-beside-action - new-skill: authority envelope in the guide's positive form; scope-beside-action - evaluate: read a tie honestly, measure the outcome not proxies, layered evaluation for side-effecting skills Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
- expose ce-skill-work to Cursor via .cursor/skills symlink; guard covers all three paths - owned procedures and required-coverage menus are protocol, not findings Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ce920218ba
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- evaluate: a missing capability is a recorded skip with a safe finish; convenience is not Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
…a phrasing list Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
… a phrasing list Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9140e88d1c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…ymlink .claude and .cursor to it Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
…s/skills directly Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
…agents/skills Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 61867efb76
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- new-skill defers to the working agreement's contribution gate instead of paraphrasing it - review-mode summary ending applies only where a summary channel exists - cursor spec notes .agents/skills discovery; AGENTS.md review-feedback rules bind any resolver Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 56642a8. Configure here.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 56642a8627
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- completion report: one statement of where review output goes (findings list is the report on findings-only callers); review-skill defers to it - new-skill: the directory exclusion is auxiliary docs, not scripts/references/assets Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a9cae51d0f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
- new-skill Done accepts a recorded capability skip, matching evaluate.md Claude-Session: https://claude.ai/code/session_015USMWYow4KB4LvG8GmGf91
* fix(ce-commit-push-pr): root PR stacks on the parent PR the user named (EveryInc#1365) * fix(ce-babysit-pr): decode gh output as UTF-8 on Windows (EveryInc#1368) * fix(ce-prototype): cover decisions settled by seeing, not just driving (EveryInc#1369) * perf(tests): cut suite wall time by splitting the largest test file (EveryInc#1370) * fix(tests): stop the cross-model routes test reading the working tree (EveryInc#1371) * fix(ce-doc-review): ask only where a real choice exists, batch the rest (EveryInc#1373) * chore(orca): re-pin upstream provenance baseline to 421a337 Co-authored-by: Cursor <cursoragent@cursor.com> * fix(ce-doc-review): align Orca ownership wording with Apply routing Co-authored-by: Cursor <cursoragent@cursor.com> * feat(ce-prototype): add a seeing-mode craft floor and durable storage (EveryInc#1374) * fix(ce-pov): stop the panel guessing the cross-model host argument (EveryInc#1375) * chore(cross-model): pin the Grok peer to 4.6 (EveryInc#1376) * docs(skills): rewrite user skill pages for accuracy and clearer use (EveryInc#1377) * fix(commit): append known plan unit ids to commit subjects (EveryInc#1379) * fix(ce-work): stop sandboxed workers committing in linked worktrees (EveryInc#1382) * fix(ce-doc-review): edit HTML plans in native format (EveryInc#1381) * fix(ce-code-review): cover adversarial after quota or auth no-review (EveryInc#1380) * fix(skills): correct a rejected dispatch instead of spending the fallback (EveryInc#1383) * fix(ce-compound): find Claude sessions started outside the repo root (EveryInc#1378) * ci(windows-native): retry peer-job-runner smoke on ctypes flake (EveryInc#1384) * fix(ce-debug): stop asking at the handoff, stop shipping unoffered work (EveryInc#1385) * docs(solutions): record why skill gates state conditions, not git commands (EveryInc#1386) * fix(skills): drop the residual-findings record file for real sinks (EveryInc#1387) * fix(ce-doc-review): run the cross-model pass when CROSS_MODEL_PEERS is unset (EveryInc#1389) * fix(ce-proof): sync with current Proof v3 contract (EveryInc#1390) * fix(skill-authoring): make goal-first the default when authoring and reviewing skills (EveryInc#1391) * fix(cross-model): let reviews run on Fable and pin model/effort from CE config (EveryInc#1392) * docs(cross-model): point superseded peer benchmarks at the luna/xhigh decision (EveryInc#1393) * fix(cross-model): discover the Codex.app-bundled codex CLI and name the peer-CLI requirement (EveryInc#1395) * feat(cross-model): add cross_model_review_mode checkout egress gate (EveryInc#1396) * fix(ce-compound-refresh): compare knowledge-track learnings against guidance they name (EveryInc#1399) * docs(solutions): capture the named-guidance contradiction-check learning (EveryInc#1400) * fix(ce-compound): prefer the repo's own frontmatter vocabulary over the Rails-era enums (EveryInc#1394) * fix(ce-work): stop asking about branches before starting work (EveryInc#1397) * fix(review): answer covered cases on skill prose with the condition, not a patch (EveryInc#1401) * fix(scratch): fall back to $TMPDIR when /tmp cannot host the scratch root (EveryInc#1398) * feat(ce-skill-work): repo-local skill for authoring, editing, reviewing, and responding to review on skills (EveryInc#1402) * fix(ce-pov): reject non-final peer positions instead of folding them in (EveryInc#1403) * feat(manifest): add Agent Plugins v1.0.0 manifest support (EveryInc#1345) * chore: release main (EveryInc#1354) * fix(ce-work): run cross-model verification on warm checkouts (EveryInc#1404) * fix(orca): reconcile upstream skill contracts * fix(orca): preserve additive Codex session roots --------- Co-authored-by: Trevin Chow <trevin@trevinchow.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
1 participant
Summary
Agents working on
skills/**in this checkout now have a procedure to run, not only a standard to remember. #1401 put the standard and the reviewer rules in AGENTS.md; this PR moves the procedures — how to author a new skill, edit an existing one, review a skill change, and act on review feedback — into a repo-local skill that loads when the work starts, and slims AGENTS.md back to what must be always-loaded plus a pointer..agents/skills/ce-skill-work/(one copy, which Codex and Cursor discover directly;.claude/skillsis a symlink to.agents/skillsfor Claude Code). Not part of the distributed plugin — plugin skills live underskills/, and the frontmatter/convention tests already skip.claudeand.agents.SKILL.md— trigger-only description; outcome / done / non-goal; the standard (guide is authority; five things every block hands the agent); rules that hold in every mode; a mode table; the completion-report contract. New rules that AGENTS.md never carried: provenance-gated removal (a line survives if a test asserts it, a learning records it, or a commit added it for a named bug — cite which; otherwise it is a cut and you don't relitigate a defended keep), for every mandate removed, name what now decides (a required gate stays), per-step done criteria, and validate-to-risk.references/new-skill.md— outcome spine before workflow, activation contract, layering order, repo inventory items, eval.references/edit-skill.md— read the block's goal, provenance search, an audit-question table (phantom handoff, step machinery, capability restatement, filler rationale, enumerated cases, prescribed mechanism, vestigial mode, duplication), bring the touched block to the standard, scope rule, repeated-repair signal, over-cut guard, exactly-once scripted edits, cross-skill contracts.references/review-skill.md— what a finding is; Change / Verify / Consider with the evidence each requires; output shape.references/respond-to-review.md— Evidence → Owning layer → Mechanism → Reconcile → Stop the accretion loop; covered case → answer with the condition; second round on a block → one restatement item; rounds counted per PR.references/evaluate.md— when an eval is required, baseline-then-compare viaskill-creator, Claude and Codex by default, forward-testing hygiene (the agent must not know it is a test; artifacts, not conclusions), read transcripts, honest tie-reading, what to record; a missing capability is a recorded skip.The references also wire in the field guide's procedures at the step where each fires (decentering step, protocol-vs-judgment disappearance test with the coverage-menu guard, authority envelope, scope-beside-action) and practices from the Anthropic and Codex
skill-creatorskills (pushier trigger-only descriptions for under-triggering, degrees of freedom by fragility, one-level references / TOC / no auxiliary files, ALWAYS/NEVER as an audit smell).AGENTS.md —
## Working on Skillsis now: the standard (3 lines), the invoke pointer, three always-on rules, the pinnedUser-Facing Skill Invocationsrule (parity-tested, so it stays), the reviewer section bots read, and a short "acting on review" paragraph pointing at the skill.Guard:
tests/repo-local-ce-skill-work.test.tspins the skill path, the.claude/skillsfolder symlink, the mode→reference mapping, the per-mode report, and the AGENTS.md pointer.Validation
bun run test3128 pass;release:validateandplugin:validateclean.One behavioral eval, Claude only (fresh subagents, same scenario: a real round-3 finding from #1397 asking to add a "missing
origin" branch to the rename probe):declined— named it a case against a stated condition and stopped.ce-skill-work:fixed-differently— recognized the second-round-on-same-block signal, folded the rounds into one restatement (condition + safe direction, exit-code enumeration deleted — the fix fix(ce-work): stop asking about branches before starting work #1397 reached four rounds later), flagged the auto-rename removal as a contract decision, wrote the record line, named the eval it would run.Not run on Codex; a true no-guidance control (outside this checkout) was not run either.
Security Disclosure
No security-relevant changes.
Agent Disclosure
Claude Code · claude-fable-5