skill-judge

To evaluate Agent Skill quality, audit the SKILL.md file against a multi-dimensional rubric covering specification compliance, frontmatter validation, and loading triggers to generate actionable improvement scores.

Validating SKILL.md frontmatter and structure requires checking YAML fields for name and description, ensuring the body is self-contained, and verifying progressive disclosure for optional references and assets.

A multi-dimensional rubric scores Agent Skill packages by analyzing eight dimensions including knowledge delta, mindset, anti-patterns, pattern recognition, and usability to identify gaps and guide targeted improvements.

You do not need specific dependencies to audit a SKILL.md file; the evaluation only requires a valid Skill unit containing YAML frontmatter and a self-contained markdown body to generate a report.

Skill units fail specification compliance checks when frontmatter lacks required name and description fields, the SKILL.md body is not self-contained, or loading triggers and progressive disclosure are improperly configured.

You can benchmark existing SKILL.md files by providing the content to an evaluator that scores freedom calibration, anti-pattern quality, and practical usability, returning concrete recommendations for targeted improvements.