Skill Audit

Audit and score OpenClaw AgentSkills with a multi-dimensional scoring system.

Quick Start

"audit this skill" or "score path/to/SKILL.md"

Language Policy

• Audit reports: Always generate in two languages:
  1. User's language — for accessibility (detect from request language or target SKILL.md body language)
  2. English — for global shareability and comparability
• Both versions should be complete (scores, dimension analysis, improvement recommendations).
• Format: write the user-language version first, then the English version, separated by a clear divider.
• If the user already requested English, a single English version suffices (no duplication).

Scoring Overview

Total (0–100) = Weighted Average

Dimension           Weight  Score Range  Description
─────────────────── ─────── ─────────── ───────────
A. Structure         20%    0–100       Spec compliance, frontmatter, file structure
B. Triggering        15%    0–100       description quality, keywords, triggering
C. Style Guide       20%    0–100       Judgment criteria, preservation/compression rules
D. Workflow          15%    0–100       Phase 0, sub-agent design
E. Sub-Agent Design  15%    0–100       Context injection, templates
F. Conciseness       15%    0–100       SKILL.md length, references separation

Grade: A(90+) | B(75+) | C(60+) | D(45+) | F(\x3C45)

How to Audit

1. Read target SKILL.md and any references/
2. Score each dimension using the rubric below
3. Calculate weighted total and grade
4. Report score card + improvement recommendations

Dimension A: Structure (20%)

Frontmatter, file structure, agentskills spec compliance.

Max: 100

Dimension B: Triggering (15%)

Whether the description triggers the agent correctly.

Max: 100

Dimension C: Style Guide (20%)

Whether judgment criteria (preservation/compression) are specified.

Applicability:

• Applies: Text processing/transformation/summarization skills (tasks that modify original content)
• Partially applies: Code generation/modification skills (may need existing code preservation criteria)
• N/A: Structured data management (JSON/DB CRUD), simple tool wrappers, configuration management

For N/A skills, exclude this dimension and redistribute weights among remaining dimensions:

• Only C is N/A: A:24%, B:18%, D:18%, E:18%, F:22%
• Only E is N/A: A:24%, B:18%, C:24%, D:18%, F:16%
• Both C+E are N/A: A:30%, B:20%, D:20%, F:30%

Max: 100

Dimension D: Workflow (15%)

Whether Phase 0, execution order, and validation steps exist.

Max: 100

Dimension E: Sub-Agent Design (15%)

Sub-agent prompt design quality.

Applicability:

• Applies: Skills that use sub-agents (parallel/sequential chunk processing, etc.)
• Partially applies: Skills that optionally mention sub-agent delegation (core is single-agent)
• N/A: Skills that do not use sub-agents at all

For N/A skills, exclude this dimension and redistribute weights among remaining dimensions:

• Only C is N/A: A:24%, B:18%, D:18%, E:18%, F:22%
• Only E is N/A: A:24%, B:18%, C:24%, D:18%, F:16%
• Both C+E are N/A: A:30%, B:20%, D:20%, F:30%

Max: 100

Dimension F: Conciseness (15%)

Context window efficiency.

Max: 100

Output Format

Delivery

Always do both:

1. File: Save the full audit report (bilingual) to ~/.openclaw/workspace/skill-audit-reports/[skill-name]-audit-[YYYY-MM-DD-HHmm].md. If the same skill is audited again, a new timestamped file is created — never overwrite previous reports. Never save inside the target skill's directory — keep skill folders clean for distribution.
2. Response: Include a summary card in your reply — total score, grade, top 3 improvement recommendations, and file path. Do NOT paste the full report in the response.

Bilingual Output

Per the Language Policy above, generate two complete versions:

1. User's language version (first)
2. English version (second)

Separate them with a clear divider:

---
## English Version

If the user requested English, a single English version suffices.

Score Card

# Skill Audit: [skill-name]

**Version**: skill-audit v1.0
**Date**: YYYY-MM-DD
**Target**: path/to/SKILL.md

## Score: 78/100 (B)

| Dimension | Score | Weight | Weighted |
|-----------|-------|--------|----------|
| A. Structure | 85 | 20% | 17.0 |
| B. Triggering | 90 | 15% | 13.5 |
| C. Style Guide | 60 | 20% | 12.0 |
| D. Workflow | 70 | 15% | 10.5 |
| E. Sub-Agent Design | 80 | 15% | 12.0 |
| F. Conciseness | 85 | 15% | 12.8 |
| **Total** | | | **77.8** |

## Grade: B

---

## Dimension Details

### A. Structure (85/100)
✅ Frontmatter exists (+15)
✅ name valid: "summarize" (+10)
✅ description present (+15)
...
❌ SKILL.md over 500 lines (-10)
→ Tip: Split detailed specs into references/

### B. Triggering (90/100)
...

## Improvement Recommendations

1. **[High]** Add preservation/compression criteria to Style Guide → C score +20 expected
2. **[Medium]** Specify Phase 0 → D score +15 expected
3. **[Low]** Add references/ TOC → A score +5 expected

Improvement Priority Labels

• [Critical] — Skill may not function properly
• [High] — Major impact on quality
• [Medium] — Meaningful score improvement if addressed
• [Low] — Minor improvement

Batch Mode

Audit multiple skills at once:

"Audit all skills in the skills/ folder"

→ Generate individual score cards per skill, then produce a comparison table:

| Skill | Total | Grade | A | B | C | D | E | F |
|-------|-------|-------|---|---|---|---|---|---|
| summarize | 82 | B+ | 93 | 88 | 88 | 77 | 78 | 65 |
| changelog | 85 | B+ | 90 | 85 | 90 | 85 | N/A | 80 |
| scaffold | 81 | B | 85 | 80 | 75 | 80 | N/A | 85 |

References

• references/scoring-rubric.md — Detailed scoring rubric with examples per dimension
• examples/ — Completed audit reports and batch comparisons for validated skills