What problem does it solve? Repositories accumulate AI-steering docs (CLAUDE.md, AGENTS.md, .cursor/rules, .windsurfrules, copilot-instructions.md, and more) that contradict each other, duplicate guidance, waste tokens, and drift in terminology. This Skill audits the entire corpus together, finds the conflicts and dead weight, and applies fixes consistently across every file. ## Core Features & Use Cases - Corpus-wide audit: Discovers every steering doc by convention, import pointer, and heuristic sweep, then reads them all before judging any one of them. - Nine-dimension checks: Token budget, duplication, no-op instructions, structural fit per harness, terminology consistency, front-loading, degrees of freedom, negation, and frontmatter validity. - Scope-aware conflict resolution: Resolves contradictions by harness precedence rules (narrower scope wins) instead of asking the user about every tie. - Use Case: Your agent keeps ignoring instructions or burning tokens on a bloated CLAUDE.md. Run this Skill to get a severity-ranked findings report, then have it apply the fixes across all steering files at once. ## Quick Start Ask the agent to audit and tune all AI agent steering docs in this repository for contradictions, duplication, and token waste.
