Sigma Foundry › AI & Developer › Free tool

CLAUDE.md Auditor AI & Developer

For teams running Claude Code, Cursor or Codex on a real repo: paste your CLAUDE.md (or AGENTS.md) and get a score, line-numbered findings, a trimmed outline and a leaner template with your project facts filled in. Nothing leaves your browser.

0 lines
0audit score · 100 = lean, current, unambiguous

Findings by section 0

Trimmed outline (what to keep, condense, move to a linked doc, or remove)

Leaner CLAUDE.md template (your detected facts pre-filled)


  

    
    
    
  





Next: brief the agent before it builds. A clean CLAUDE.md sets the house rules. Commander's Intent for Coding Agents adds a briefing for each feature: four gates before the agent writes code, then a revert test, a rules judge and a comprehension checkpoint after. $37 once.
See Commander's Intent →


Want a human to do the trim? The CLAUDE.md Audit service rewrites your instructions file with you in one working session. And once the agent is running well, AgentPulse tells you the moment a long loop stalls, errors or goes quiet.
Open AgentPulse →Ask about the CLAUDE.md Audit service →



Heuristics, not a compiler: the 200-line guidance, index-over-encyclopedia and outcome-over-procedure rules come from published Claude Code and agent-instruction practice; every finding names its line so you can judge it. Rules verified 2026-09-15; tell us if something changed. Privacy · nothing you paste leaves your browser unless you email yourself a result. Keys found by the secrets check are masked everywhere in the output.