quality-audit（质量审核）

Quality Audit Skill

Systematic framework for evaluating skill quality across four dimensions: Clarity, Completeness, Accuracy, and Usefulness.

When to Use This Skill

• Reviewing a new skill before adding to the registry
• Auditing existing skills for quality improvements
• Creating quality rubrics for skill validation
• Standardizing skill quality across the library
• Preparing skills for production use

Core Principles

The Four Quality Dimensions

Scoring Scale (1-5)

Passing Criteria

• Minimum: 3.0 weighted average (acceptable)
• Target: 4.0 weighted average (good)
• Exceptional: 4.5+ weighted average (excellent)
• Blocking: Accuracy must be ≥3.0 (no dangerous advice)

Audit Workflow

Phase 1: Structure Check

checklist:
  structure:
    - [ ] Has valid YAML frontmatter
    - [ ] Contains required metadata (name, description)
    - [ ] Follows progressive disclosure (Tier 1 → 2 → 3)
    - [ ] Sections are logically ordered
    - [ ] Token estimate is reasonable (<5000 for core)

Phase 2: Content Evaluation

checklist:
  content:
    - [ ] "When to Use" section is clear
    - [ ] Core principles are well-defined
    - [ ] Code examples are complete and runnable
    - [ ] Anti-patterns are documented
    - [ ] Troubleshooting guidance exists

Phase 3: Dimension Scoring

For each dimension, evaluate against specific criteria:

Clarity Criteria:

• Well-organized sections with logical flow
• Concise explanations without jargon overload
• Code examples are readable and well-commented
• Progressive disclosure from simple to complex

Completeness Criteria:

• Covers core concepts thoroughly
• Includes edge cases and error handling
• Provides both do's and don'ts
• Has working examples for main use cases

Accuracy Criteria:

• Code examples compile/run without errors
• Follows current best practices (not deprecated)
• Security considerations are correct
• Performance claims are verifiable

Usefulness Criteria:

• Examples solve real-world problems
• Can be applied immediately
• Scales to production use cases
• Includes troubleshooting guidance

Phase 4: Report Generation

## Audit Report: {skill_name}

**Date**: {date}
**Auditor**: {auditor}
**Status**: {PASS|FAIL|NEEDS_REVIEW}

### Scores

| Dimension | Score | Weight | Weighted |
|-----------|-------|--------|----------|
| Clarity | {x}/5 | 25% | {x*0.25} |
| Completeness | {x}/5 | 25% | {x*0.25} |
| Accuracy | {x}/5 | 30% | {x*0.30} |
| Usefulness | {x}/5 | 20% | {x*0.20} |
| **Total** | | | **{sum}/5** |

### Issues Found

- [CRITICAL] {issue description}
- [MAJOR] {issue description}
- [MINOR] {issue description}

### Recommendations

1. {actionable recommendation}
2. {actionable recommendation}

Implementation Patterns

Pattern 1: Quick Audit (5-minute review)

Use for rapid assessment of skill quality:

# Run automated structure checks
cortex skills audit <skill-name> --quick

# Output: Pass/Fail with basic metrics

Quick Audit Checks:

1. YAML frontmatter valid?
2. Required sections present?
3. Code blocks have language tags?
4. No TODO/FIXME markers?
5. Token count reasonable?

Pattern 2: Full Audit (15-30 minute review)

Comprehensive evaluation with human review:

# Generate full audit report
cortex skills audit <skill-name> --full

# Interactive mode for scoring
cortex skills audit <skill-name> --interactive

Full Audit Process:

1. Run automated checks
2. Read through content manually
3. Test code examples
4. Score each dimension
5. Document issues and recommendations
6. Generate report

Pattern 3: Comparative Audit

Compare skill against reference implementation:

# Compare against template-skill-enhanced
cortex skills audit <skill-name> --compare template-skill-enhanced

Pattern 4: Batch Audit

Audit multiple skills for registry health:

# Audit all skills in a category
cortex skills audit --category security

# Audit skills below threshold
cortex skills audit --below-score 3.5

CLI Commands

# Basic audit
cortex skills audit <skill-name>

# Options
  --quick           Quick structural check only
  --full            Full audit with all dimensions
  --interactive     Interactive scoring mode
  --output FILE     Write report to file
  --format FORMAT   Output format (markdown|json|yaml)
  --compare SKILL   Compare against reference skill
  --fix             Auto-fix simple issues (formatting)

Creating Custom Rubrics

Skills can define custom rubrics in validation/rubric.yaml:

# validation/rubric.yaml
version: "1.0.0"
skill_name: my-skill

dimensions:
  clarity:
    weight: 25
    criteria:
      - "API examples use realistic data"
      - "Error handling is shown for each operation"
  completeness:
    weight: 25
    criteria:
      - "Covers all HTTP methods"
      - "Includes pagination patterns"
  accuracy:
    weight: 30
    criteria:
      - "Follows REST conventions"
      - "Security headers documented"
  usefulness:
    weight: 20
    criteria:
      - "Examples work with common frameworks"

passing_criteria:
  minimum_score: 3.5  # Higher bar for this skill
  required_dimensions:
    - accuracy
    - completeness

Best Practices

Do

• Be specific - "Line 45: SQL query vulnerable to injection" not "has security issues"
• Be actionable - Include how to fix each issue
• Be fair - Use the same standards consistently
• Document evidence - Quote specific content for each score
• Prioritize - Critical issues first, suggestions last

Don't

• Score based on personal style preferences
• Mark deprecated patterns without suggesting alternatives
• Fail skills for missing optional sections
• Ignore security issues regardless of other scores
• Rush through audits for complex skills

Anti-Patterns

The Rubber Stamp

Problem: Approving skills without thorough review Why it's bad: Low-quality skills erode trust in the library Fix: Use the full audit checklist, test code examples

The Perfectionist Block

Problem: Failing skills for minor issues Why it's bad: Prevents useful skills from being available Fix: Distinguish between blocking issues and suggestions

Score Inflation

Problem: Giving high scores without justification Why it's bad: Makes scores meaningless Fix: Document specific evidence for each score

Integration with CI/CD

# .github/workflows/skill-quality.yml
name: Skill Quality Gate

on:
  pull_request:
    paths:
      - 'skills/**'

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Install cortex
        run: pip install cortex
      - name: Audit changed skills
        run: |
          for skill in $(git diff --name-only HEAD~1 | grep 'skills/' | cut -d'/' -f2 | uniq); do
            cortex skills audit "$skill" --quick --fail-under 3.0
          done

Troubleshooting

"Audit fails but skill looks fine"

1. Check YAML frontmatter syntax
2. Verify all required sections exist
3. Ensure code blocks have language tags
4. Check for hidden characters (copy/paste issues)

"Scores seem inconsistent"

1. Review the scoring guide for each dimension
2. Calibrate by auditing template-skill-enhanced first
3. Use --interactive mode for clearer criteria

External Resources

• Skill Template Reference
• Rubric Schema
• Skill Creator Guide

Changelog

1.0.0 (2026-01-05)

• Initial release
• Four-dimension scoring framework
• CLI integration
• CI/CD workflow example