What problem does it solve? AI rules files (CLAUDE.md, AGENTS.md, .cursor/rules/*.mdc, copilot instructions) silently degrade over time: they grow too long, accumulate vague advice, duplicate knowledge the model already has, and bury critical rules where the model pays least attention. This Skill audits any rule-for-AI markdown file and turns that gut feeling of "my agent ignores my rules" into a scored, actionable report. ## Core Features & Use Cases - Five-check scorecard: Scores length, direct code snippets, vague language, redundant knowledge, and rule ordering (U-shaped attention), each with OK/WARN/FAIL verdicts and line-referenced findings. - Grounded rewrites: Every vague phrase ("write clean code") gets a concrete, testable replacement derived from the project's own stack and conventions, never generic advice. - Interactive reordering: Proposes a better section order and, only with explicit user approval via AskUserQuestion, moves sections while preserving all rule content. - Use Case: Run it on a 600-line CLAUDE.md that your agent keeps ignoring; get a scorecard showing 3 vague phrases, 4 redundant paragraphs, critical security rules buried at line 312, and a top-3 action list to fix it. ## Quick Start Ask the assistant to review your AI rules file, for example: run /10x-rule-review on CLAUDE.md and show me the scorecard.
