skill-judge

To evaluate a SKILL.md file, load the target skill and follow an evaluation protocol that scores frontmatter, body content, loading triggers, and references across eight dimensions to generate a structured report with actionable recommendations.

The best way to identify gaps in agent skill design is by quantifying the knowledge delta between the skill and expert-level expectations, applying a 120-point scoring rubric across eight dimensions including specification compliance and practical usability.

Multidimensional skill evaluation scoring works by assessing eight specific dimensions totaling 120 points, including anti-pattern quality, progressive disclosure, and pattern recognition, to measure compliance and identify areas for improvement.

Yes, you can use a structured evaluation protocol to audit existing agent skills by gathering evidence, scoring multidimensional criteria, and producing a formal Skill Evaluation Report to guide improvements and reduce token waste.

A limitation of evaluating agent skills with a rubric is that it primarily targets SKILL.md frontmatter, body content, loading triggers, and references, meaning it assesses design specifications rather than runtime execution performance.

You need to assess specification compliance for skill packages when auditing existing files, improving skill quality, or teaching new evaluators how to measure design completeness against official best practices.