skill-judge

# skill-judge

Meta-skill for evaluating skill quality before publishing or during audits.

## When to Use

| Trigger | Action |
| ------- | ------ |
| Before publishing new skill | Run full evaluation |
| Reviewing skill PR | Run full evaluation |
| Auditing existing skills | Batch evaluation |
| "Is this skill worth it?" | Run Knowledge Delta check |

## Quick Assessment

Before deep evaluation, answer:

1. **Does Claude already do this well?** If yes, skill is likely [R]edundant
2. **Would a 3-line prompt achieve the same?** If yes, too little delta
3. **Is the problem actually bounded?** If no, redesign first

## Evaluation Framework (120 points)

| Dimension | Points | Question |
| --------- | ------ | -------- |
| Problem Definition | 15 | Is the problem clear and bounded? |
| Knowledge Delta | 20 | Does it add [E]xpert or [A]ctivation knowledge? |
| Description Quality | 15 | WHAT + WHEN + KEYWORDS present? |
| Progressive Disclosure | 15 | Right level of detail per section? |
| Freedom Calibration | 15 | Constraints match task nature? |
| Output Format | 10 | Clear templates/structure? |
| Anti-Pattern Coverage | 15 | Common mistakes addressed? |
| Testability | 15 | Can we verify it works? |

**Scoring guide:** See [evaluation-dimensions.md](references/evaluation-dimensions.md)

## Score Interpretation

| Score | Verdict | Action |
| ----- | ------- | ------ |
| 100-120 | Excellent | Ship |
| 80-99 | Good | Ship with minor tweaks |
| 60-79 | Needs work | Revise before publishing |
| 40-59 | Significant issues | Major redesign needed |
| < 40 | Fundamental problems | Reconsider if skill is needed |

## Knowledge Delta Classification

Every skill must justify its existence via knowledge delta.

| Type | Symbol | Definition | Example |
| ---- | ------ | ---------- | ------- |
| Expert | [E] | Domain expertise Claude lacks | Security threat models, industry-specific regulations |
| Activation | [A] | Knowledge Claude has but needs prompting | Structured framewo