10x-rule-review

Pass the file path as an argument, such as @CLAUDE.md, AGENTS.md, or .cursor/rules/api.mdc. The skill reads the file, runs five checks, and prints a scorecard with verdicts and concrete fixes for each finding.

It runs five checks: file length, direct code or config snippets, vague untestable language, redundant knowledge the model already knows, and rule ordering. Each check returns an OK, WARN, or FAIL verdict with line-referenced findings.

Yes, it treats .mdc files like any other rules-for-AI markdown. Frontmatter such as globs and alwaysApply is treated as configuration, and rule lines are counted from after the frontmatter block.

No, the default output is read-only. The only possible edit is a section reorder from Check 5, and that happens only after you explicitly approve the proposed order through an interactive prompt.

Models pay more attention to the start and end of long contexts, so critical rules buried in the middle are statistically less likely to be followed. The review flags files where hard rules appear after line 200 or where intro content dominates the top.

It scores the condition of the rule artifact, not the project's architecture or tech choices. It never rewrites rule content, generates a fixed version of the file, or merges scorecards when multiple files are reviewed.