A 12-Point CLAUDE.md Audit You Can Run in 20 Minutes

An audit is not a vibe check. Each of these twelve points has a test you can run in under two minutes, and a clear pass condition.

Score one point per pass, half per warning. 10+ is usable, 6–9 is fragile, under 6 will produce silent failures.

1. One file, one owner

Test: search the repo for other copies. find . -name "CLAUDE.md" -o -name "AGENTS.md" Pass: one authority file, one named owner, no per-repo forks.

2. Scope boundary

Test: count prohibitions. grep -ciE "never|do not|must not" CLAUDE.md Pass: at least three, each naming a concrete action.

3. Gate list

Test: search for the five money actions: post, email, price, refund, deploy. Pass: all five appear with a rule and who may pass them.

4. Definition of done

Test: search grep -i "done" CLAUDE.md Pass: a sentence that defines done as artifact plus check plus next step.

5. Status format

Test: search for the three labels. Pass: done / open / next appears with an example.

6. Secrets hygiene

Test: grep -iE "key|token|secret|\\.env" CLAUDE.md Pass: the rule says variable names only, and stops before printing env values.

7. Precedence

Test: search for conflict or priority language. Pass: an explicit order, e.g. safety > money > speed.

8. Question policy

Test: search for "ask". Pass: a limit (one question) and a rule for when it is allowed.

9. Output path

Test: does the file say where work lands? Pass: a folder or naming convention, not "the user's workspace".

10. Context budget

Test: wc -w CLAUDE.md Pass: under 700 words. Over 1,200 is a fail: it costs tokens on every task.

11. No stale placeholders

Test: grep -nE "\\{\\{|TODO|TBD" CLAUDE.md Pass: none, or a dated note saying what fills them.

12. Honesty rule

Test: search for invented numbers. Pass: a line banning fabricated metrics, reviews, customers and revenue.

What the scores mean in practice

Teams usually fail 4, 5, 8 and 10. That combination is exactly the "agent says done and nothing changed" pattern: no definition of done, no status format, unlimited questions, and a file so long nobody reads it.

Fix those four first. Most teams gain more from that than from any new tool.

Run it automatically

The free score tool runs these checks on a pasted file and names the three to fix first. For a team, the $299 audit adds line-by-line scoring across every member's file, one corrected version, a delete list and a recording. If you only need one file fixed, the $29 context fix returns it in 24 hours.

FAQ

How often should I re-run the audit?

Every time a rule changes, and once a quarter. Otherwise the file drifts back to prose.

Can I audit a file I did not write?

Yes. The checks are about structure and specificity, not style.

What does the pack cost?

$9.99 per month at the founding price, or $79 per year. It moves to $19.99 on 2026-12-01. Details on the pricing page.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {"@type": "Question", "name": "How often should I re-run the audit?", "acceptedAnswer": {"@type": "Answer", "text": "Every time a rule changes, and once a quarter. Otherwise the file drifts back to prose."}},
    {"@type": "Question", "name": "Can I audit a file I did not write?", "acceptedAnswer": {"@type": "Answer", "text": "Yes. The checks are about structure and specificity, not style."}},
    {"@type": "Question", "name": "What does the pack cost?", "acceptedAnswer": {"@type": "Answer", "text": "9.99 USD per month at the founding price, or 79 USD per year. It moves to 19.99 USD on 2026-12-01. Details: https://xynex.sekai.center/pricing"}}
  ]
}

All guides · 中文版：20 分鐘跑完的 CLAUDE.md 12 項檢查