skill-judge

Run the skill-judge evaluation protocol: read the SKILL.md completely, classify each section as Expert, Activation, or Redundant knowledge, then score 8 dimensions totaling 120 points. The output is a graded report with critical issues and top improvements.

A good description answers three questions: WHAT the skill does, WHEN it should be used, and which KEYWORDS trigger it. It must list specific capabilities, include explicit trigger scenarios like "Use when...", and contain searchable terms so the agent activates it correctly.

The total is the sum of 8 dimension scores out of 120 points, converted to a percentage. Grades are A (>=90%), B (>=82.5%), C (>=70%), D (>=60%), and F (<60%). Scores below 70% trigger mandatory failure-pattern diagnosis.

Nine named patterns exist, including The Tutorial (explaining basics the model knows), The Dump (800+ line SKILL.md with no progressive disclosure), The Invisible Skill (vague description that never triggers), and The Freedom Mismatch (rigid scripts for creative tasks).

The quick checklist in references/checklist.md is intended for fast audits, not full evaluations. The full 8-dimension protocol is for thorough scoring; loading the checklist during a full evaluation is explicitly discouraged.