keel-audit
Skill9 starsChanged 51 days agoMITView on GitHub
Periodic rules-compliance spot-check — spawns the read-only auditor subagent over the commits since the last audit, reports violations with file:line + rule §, fixes only what the user approves. Skips itself on an empty (phase-0) project.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- AWenSu/keeltheir own version, first seen Sep 19, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- AWenSu/keel★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- keel-adoptSkill
- keel-agent-team-createSkill
- keel-agent-team-startSkill
- keel-autopilotSkill
- keel-compactSkill
- keel-continueSkill
- keel-distillSkill
- keel-handoverSkill
- keel-phase-reviewSkill
- keel-pilotSkill
- keel-planSkill
- keel-researchSkill
- keel-statsSkill
- keel-teamSkill
- keel-tidySkill
- keel-updateSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
When: the SessionStart hook nudges (default: >25 commits since the last audit), at /keel-phase-review, or on demand.
- Very specific: line 29
Record: Write the findings table + the follow-up table + the auditor's own hit rate (refuted claims / total claims, from your spot-check in …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/keel-audit — check the recent work against rules.md
When: the SessionStart hook nudges (default: >25 commits since the last audit), at /keel-phase-review, or
on demand. Complements — never replaces — /keel-distill (memory lint) and verifier (single claims): this
is the "is the project still following its own constitution?" sweep. Post-compact drift gets a free
30-second inline self-check from the re-ground hook; this skill is the deep pass.
- Follow-up FIRST — what became of the LAST audit's findings. Open the previous report
(
reports/<date>-audit.md, newest) and, before any new finding, write one line per earlier finding:id · decision (fixed / deferred / rejected) · landed? (sha or "not landed") · measurement then → now. Count the fourth outcome separately — silently dropped (evaluated, never tracked) — and re-measure every rejection's stated reason: is it still true today? A landed fix whose number did not move is itself a finding. First audit: write "first round, no follow-up" — the section is never skipped. Field: an audit's 30 recommendations were all evaluated and none tracked until this table existed; a finding-producing audit that never measures change is noise. - Scope. Phase-0 guard: if the source tree is still empty and the docs are placeholders, report
"phase 0 — nothing to audit yet" and stop. Otherwise the range is
$(cat .claude/last-audit)..HEADwhen the marker exists (else the last ~25 commits). - Run. Spawn the
auditorsubagent (.claude/agents/auditor.md) with the range — don't audit inline; the point is fresh eyes in an isolated context. Spot-check its claims before acting (§4.11: verify, don't trust). - Report & fix. Present the severity-ranked table. Apply ONLY user-approved fixes; real-but-deferred
findings get one line each in
TASKS.md ## Discovered. Never fix silently (§10.36). - Record. Write the findings table + the follow-up table + the auditor's own hit rate
(
refuted claims / total claims, from your spot-check in step 2 — a critic whose accuracy is never measured becomes a voice nobody can challenge) toreports/<date>-audit.md— that file is what the NEXT audit's step 0 reads. Write the audited HEAD sha to.claude/last-audit(resets the hook's due-counter); add the HANDOVER block (a) one-lineraudit @ <sha>: N findings, M fixed, hit rate K/N; commit with approval (rules.md §1.3). - Not this skill's job: critiquing whether a RULE is right or the process is healthy. That is the
owner-triggered
observeragent (.claude/agents/observer.md) — compliance here, critique there; neither output stands in for the other's.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.