skill-review

Skill Review is a meta-skill for Claude Code that audits the quality of other skills against a documented set of best practices. It solves a concrete authoring problem: skill authors rarely have an objective yardstick for whether a SKILL.md is well-structured, discoverable, and token-efficient, so this skill turns that judgment into a repeatable review process. Its guiding 'Iron Law' is to be specific and actionable, never offering vague feedback without stating exactly what to change and why it matters for model output quality.

The skill drives a three-step workflow: load the target skill and inventory its files, analyze it across five dimensions (structure compliance, description quality, workflow design, token efficiency, and anti-pattern detection), and produce a report of strengths followed by prioritized suggestions. A companion reference file, review-criteria.md, supplies detailed checks for each dimension — such as the 500-line SKILL.md limit, keyword-bombing indicators for descriptions, confirmation-gate requirements before destructive operations, progressive-loading signals, and anti-patterns like vague directives or placeholder residue. Every suggestion is expected to state what the issue is, where it lives, and the concrete fix.

It is aimed at skill authors, prompt engineers, and teams building libraries of Claude Code skills who want consistent, opinionated quality reviews. Typical uses include linting a newly written skill, auditing an existing one before publishing, or getting a prioritized improvement list. It is read-only guidance that analyzes files and reports findings; it performs no destructive actions.