Rate Skill

Audits a single SKILL.md against current activation-driven authoring practice and returns a letter grade, weighted category scores, prioritized findings with concrete patches, named strengths, and a projected grade after fixes.

The rubric is anchored to (a) the agentskills.io / Claude Code frontmatter spec, (b) Anthropic's skill-creator guidance, and (c) Seleznov's activation study (n=650, p<0.0001) showing directive third-person descriptions activate ~20× more reliably than passive prose.

Workflow

1. Resolve the input path. If the user passes a directory, look for SKILL.md inside it. If nothing is passed, ask once: "Which SKILL.md should I rate?"
2. Read the whole file. Parse frontmatter and body separately. Count body lines (exclude frontmatter).
3. Detect the skill type — methodology (phases + checklists), reference (schemas + tables), generator (produces an artifact), or auditor (reviews an artifact). State it in the report.
4. Score each category 0–100 against the rubric below, weight, sum, map to a letter.
5. Emit the report in the shape under Output Format.
6. Estimate the post-fix grade assuming the P0 and P1 findings are applied.

Rubric (weights sum to 100)