claude-md-audit
Skill1 starsChanged 12 days agoMITView on GitHub
Audit CLAUDE.md files across one or many repos for stale information — dead path references, obsolete workspace roots, old model names, drifted file maps. Use when asked to "audit CLAUDE.md", "check for stale docs/instructions", or after restructuring a repo.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- spongetimblin/claude-skill-librarytheir own version, first seen Oct 25, 2026
- timothywarner-org/claude-codetheir own version, first seen Jul 20, 2026
- illinigirl/claude-code-toolkittheir own version, first seen Jun 23, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- timothywarner-org/claude-code★ 223same name
- spongetimblin/claude-skill-library★ 1same name
- illinigirl/claude-code-toolkit★ 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 repository28 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- cloudflare-oneSkill
- cloudflareSkill
- component-factorySkill
- concurrency-probeSkill
- core-sampleSkill
- corroborateSkill
- decision-telemetrySkill
- disk-auditSkill
- engramSkill
- frontend-designSkill
- git-forensicsSkill
- high-signal-engineSkill
- local-swarmSkill
- model-inventorySkill
- muster-triageSkill
- precise-coding-assistantSkill
- preflightSkill
- pr-triageSkill
- read-arxiv-paperSkill
- run-mazze-leczzare-blogSkill
- session-journalSkill
- shipSkill
- stratum-project-initSkill
- touchstoneSkill
- transcript-handoffSkill
- web-perfSkill
- workspace-syncSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
0
Distinctive
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 16
Run scan.py from this skill's directory against the workspace root (it ships in the workspace plugin, so it is not under ~/.claude/skills/):
- Very specific: line 62
Write docs/audits/claude-md-audit-<date>.md in the container: a Critical section, a per-repo verdict table (STALE/clean + evidence lines), …
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/claude-md-audit
CLAUDE.md files drift: repos get restructured, machines get replaced, and the instructions keep describing a world that no longer exists. Drifted agent instructions are worse than none — agents obey them confidently.
Method (two lanes, budget-aware)
Lane 1 — deterministic (cloud Claude or plain script; this is the verdict lane):
Run scan.py from this skill's directory against the workspace root (it ships
in the workspace plugin, so it is not under ~/.claude/skills/):
python3 ~/Projects/skills/claude-code-skills/skills/claude-md-audit/scan.py ~/ProjectsPass a single project directory instead to scan just that project. The
claude-md-drift-on-start.sh SessionStart hook does this automatically for
the project each new session opens in, so most drift arrives already flagged.
The hook uses --json (structured findings) and never parses the text form.
The scanned repository is treated as untrusted. Only regular, non-symlinked
files that stay inside the root are read, capped at 1 MiB, with bad encoding
replaced rather than fatal. Anything else is reported as skipped. Paths are
escaped, so a file or directory name can't forge a finding line. Regression
tests: python3 -m unittest discover -s tests -v from this directory.
Dead-path hits leave out tokens that can't be claims about the repo's
filesystem: those with no file extension, no trailing slash, and a first
segment that isn't an existing directory (owner/repo slugs, branch names,
linux/amd64, try/catch), plus domain refs (ghcr.io/…) and
slash-separated enums.
It flags, per CLAUDE.md:
- backticked path references that don't exist on disk (verify each hit —
shorthand like
blog/PullQuote.astroand enums likeplan/implement/revieware false positives) - stale workspace roots (
🚀 PROJECTS, submodule-era language) - old model-generation strings (claude-3, sonnet-3, …)
- last-updated dates
Lane 2 — on-device summaries (Ollama, zero budget): POST each file to
http://localhost:11434/api/generate (model gemma4 or better) asking for a
2-3 sentence summary + anything time-sensitive. Summaries give reviewers
orientation; they are NOT verdicts — a small model's staleness opinions are
unreliable. Never let Lane 2 output change a verdict without Lane 1 evidence.
Always check first
~/.claude/CLAUDE.md exists. Every repo defers postures/hard-stops to it;
on a fresh machine it silently vanishes and 90% of the posture goes with it
(this happened — see the tessera-claude-anchor record).
Output
Write docs/audits/claude-md-audit-<date>.md in the container: a Critical
section, a per-repo verdict table (STALE/clean + evidence lines), and a
recommended-fixes list. Flag, don't auto-rewrite: each CLAUDE.md rewrite
changes agent behavior in that repo and deserves review in that repo's context.
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.