CLAUDE.md auditor
How it works

Score your CLAUDE.md in 60 seconds.

Six axes, one number, three concrete fixes. Built by working engineers — not a chatbot. Bring your own Anthropic API key; we never store either it or the file.

How it works

  1. You paste your CLAUDE.md and your Anthropic API key. Both go straight to our Worker, which forwards the file to Anthropic with the auditor prompt. We don't log, persist, or proxy either anywhere else.
  2. The model scores you on six axes: Specificity, Coverage, Brevity, Currency, Quirks captured, Tone fit. Each axis is 0–10 with justification. Overall is a weighted score out of 100 (Specificity and Quirks count most).
  3. You get three concrete fixes — not "consider adding more detail," but specific edits with example replacement text. Plus a one-sentence verdict suitable for a share card.

Hard caps: hardcoding a retired Claude model name caps Currency at 5. The phrase "Claude is an AI assistant created by Anthropic" caps Specificity at 4. The auditor is calibrated against six real-world cases ranging from generic-template junk (baseline 18) to mature production files (baseline 78).

Want to fix it, not just score it?

The Auditor is free. The CLAUDE.md Generator — part of the Claude Code Power Pack — does the rewrite for you. It introspects your repo, drafts a fresh CLAUDE.md, self-audits against the same rubric, and iterates once if it falls short. Calibrated against the same six cases: scores 80–85 on all of them.

Power Pack ships when its second skill clears its acceptance bar. Drop your email below if you want a heads-up — no other emails, ever.