RuleScan
ESLint for your CLAUDE.md — catch contradictions before your agent does
npx -y github:Isco0819/rulescannerEver run into this?
Like ESLint (25k stars, billions of weekly downloads) but there is zero equivalent for AI agent config files. CLAUDE.md and system prompts grow to thousands of tokens with no quality gate — contradictions, dead rules referencing removed workflows, and ambiguous instructions silently break agent behavior in ways that are nearly impossible to debug after the fact.
The fix
npx CLI that statically analyzes CLAUDE.md / .cursorrules / system prompt files, detects contradictions, dead references, and ambiguity hotspots, and outputs ESLint-style findings with line numbers and severity levels.
What it does
- contradiction detection: flags rules that directly conflict (e.g. 'always confirm before sending' vs 'send without asking') with both line refs
- dead-rule scanner: cross-checks referenced files, tools, scripts, and workflows against your actual repo tree via glob — flags anything that no longer exists
- ambiguity score: ranks every rule by interpretation-count 1-10, surfaces the top offenders so you know exactly what to rewrite before your next deploy