Skill Judge

Evaluate Agent Skills against official specifications and best practices using an 8-dimensional scoring framework (120 points total).

Decision Tree

Choose evaluation approach based on context:

Evaluation Context	Primary Focus	When to Use
Quick review (5-10 min)	Description + Knowledge Delta	Initial screening, triage
Full audit (20-30 min)	All 8 dimensions	Comprehensive quality assessment
Compliance check (5 min)	Frontmatter + Description	Format validation only
Improvement guidance (30+ min)	All dimensions + detailed feedback	Skill optimization

Workflow

Step 1: Load Evaluation Framework

MANDATORY - READ ENTIRE FILE: Before proceeding, you MUST read evaluation-guide.md completely from start to finish. NEVER set any range limits when reading this file.

Step 2: Quick Scan (5 minutes)

Read SKILL.md completely and identify:

• Skill type: Mindset (~50 lines), Navigation (~30 lines), Philosophy (~150 lines), Process (~200 lines), Tool (~300 lines)
• Line count: Is it appropriate for the type?
• Description quality: Does it have WHAT, WHEN, and keywords?
• Knowledge delta: Any obvious "explaining basics" sections?

Step 3: Dimension Evaluation (15-20 minutes)

Evaluate each dimension in order:

Priority	Dimension	Points	Why This Order
1	D4: Specification Compliance (Description)	15	Poor description = skill never used
2	D1: Knowledge Delta	20	Core dimension - determines value
3	D7: Pattern Recognition	10	Sets expectations for structure
4	D5: Progressive Disclosure	15	Checks if references are used properly
5	D2: Mindset + Procedures	15	Evaluates thinking patterns
6	D3: Anti-Pattern Quality	15	Checks for NEVER lists
7	D6: Freedom Calibration	15	Matches freedom to task fragility
8	D8: Practical Usability	15	Can Agent actually use it?

Step 4: Score Calculation

Sum all dimension scores (max 120 points). Calculate percentage and assign grade:

Score Range	Grade	Interpretation
96-120	A	Excellent - Production ready
84-95	B	Good - Minor improvements needed
72-83	C	Acceptable - Moderate improvements needed
60-71	D	Poor - Significant improvements needed
<60	F	Fail - Major redesign required

Step 5: Generate Report

MANDATORY - READ ENTIRE FILE: Before generating report, you MUST read scoring-guide.md completely.

Output structured report in this format:

# Skill Evaluation Report

## Overview
- **Skill**: [skill-name]
- **Type**: [Mindset/Navigation/Philosophy/Process/Tool]
- **Total Score**: [X]/120 ([X]%)
- **Grade**: [A/B/C/D/F]

## Dimension Scores

| Dimension | Score | Max | Notes |
|-----------|-------|-----|-------|
| D1: Knowledge Delta | [X] | 20 | [brief notes] |
| D2: Mindset + Procedures | [X] | 15 | [brief notes] |
| D3: Anti-Pattern Quality | [X] | 15 | [brief notes] |
| D4: Specification Compliance | [X] | 15 | [brief notes] |
| D5: Progressive Disclosure | [X] | 15 | [brief notes] |
| D6: Freedom Calibration | [X] | 15 | [brief notes] |
| D7: Pattern Recognition | [X] | 10 | [brief notes] |
| D8: Practical Usability | [X] | 15 | [brief notes] |

## Critical Issues (Must Fix)
1. [Issue 1]
2. [Issue 2]

## Improvement Suggestions (Should Fix)
1. [Suggestion 1]
2. [Suggestion 2]

## Strengths (Keep)
1. [Strength 1]
2. [Strength 2]

NEVER Do When Evaluating

Scoring Mistakes

• NEVER give high scores for "professional formatting" alone - content matters most
• NEVER ignore token waste - every redundant paragraph = deduction
• NEVER let length impress you - 43-line skill can outperform 500-line skill
• NEVER assume all procedures are valuable - distinguish domain-specific from generic

Evaluation Mistakes

• NEVER skip mentally testing decision trees - do they lead to correct choices?
• NEVER forgive explaining basics with "but it provides helpful context"
• NEVER overlook missing anti-patterns - no NEVER list = significant gap
• NEVER undervalue description field - poor description = skill never used

Reporting Mistakes

• NEVER give vague feedback like "improve quality" - be specific
• NEVER suggest changes without explaining WHY
• NEVER provide scores without actionable improvement suggestions

Quick Reference

Knowledge Delta Red Flags (D1)

• "What is [basic concept]" sections
• Step-by-step tutorials for standard operations
• Explaining how to use common libraries
• Generic best practices ("write clean code")
• Definitions of industry-standard terms

Knowledge Delta Green Flags (D1)

• Decision trees for non-obvious choices
• Trade-offs only experts know
• Edge cases from real-world experience
• "NEVER do X because [non-obvious reason]"
• Domain-specific thinking frameworks

Anti-Pattern Quality (D3)

• Score 0-3: No anti-patterns mentioned
• Score 4-7: Generic warnings ("avoid errors")
• Score 8-11: Specific NEVER list with some reasoning
• Score 12-15: Expert-grade anti-patterns with WHY

Description Quality (D4)

• Must answer: WHAT (functionality), WHEN (trigger scenarios), KEYWORDS (searchable terms)
• Poor: "处理文档相关功能" (vague, no triggers, no keywords)
• Excellent: "Comprehensive document creation, editing, and analysis. Use when Claude needs to work with professional documents (.docx files) for: (1) Creating new documents, (2) Modifying content, (3) Working with tracked changes"

Pattern Recognition (D7)

Pattern	~Lines	When to Use
Mindset	~50	Creative tasks requiring taste
Navigation	~30	Multiple distinct sub-scenarios
Philosophy	~150	Art/creation requiring originality
Process	~200	Complex multi-step projects
Tool	~300	Precise operations on specific formats

Freedom Calibration (D6)

• High freedom: Creative/Design tasks (frontend-design)
• Medium freedom: Code review, judgment-based tasks
• Low freedom: File format operations (docx, pdf, xlsx)

Output Format

Always output evaluation report in the structured format shown in Step 5. Include:

• Overview with total score and grade
• Dimension scores table with notes
• Critical issues (must fix)
• Improvement suggestions (should fix)
• Strengths (keep)

Do NOT output:

• Unstructured feedback
• Scores without explanations
• Generic comments without specific examples