53-item, 10-layer audit model baked into an 8-stage gated creation workflow. Not abstract advice — battle-tested rules traced to real incidents. Local execution, zero subscription.
Professional-grade skill quality assurance, zero subscription, fully open source.
10 layers covering document structure, architecture consistency, code quality, cross-file alignment, security compliance, robustness, and content quality. 27 items auto-scannable.
L1 trigger layer always loaded (~100 words). L2 workflow loaded on activation (≤300 lines). L3 references loaded on demand. Controls context window bloat by design.
34 anti-patterns, each traced to a real incident, linked to a design principle, and paired with a concrete fix. Battle-tested corrections, not abstract advice.
skill-audit.sh performs batch scanning with a single command. Checks structure, security risks, line count, code quality, and more — before every publish.
Built-in quantitative signals (stale_count, time limits, failure thresholds) prevent infinite loops and silent failures in iterative or scheduled skills.
Hard rules against credential leakage, personal data exposure, and restricted system access. Automated scanning catches violations before they ship.
Context-aware loading: only what's needed, when it's needed. Controls context window bloat by design.
Every workflow has explicit entry/exit criteria. No skipping — quality is structural, not optional.
Full 8-stage gated workflow: requirements → planning → L1/L2/L3 authoring → audit → publish.
53-item checklist across 10 layers. Automated shell scan + human-judgment items. Every rule traces to a real incident.
AP-1 through AP-34 catalog scanned automatically. Each flagged with root cause and concrete fix.
Trigger word analysis: banned-word scan, coverage test, pushy-principle check. 10+10 trigger accuracy test.
3-layer model applied: keep L2 ≤ 300 lines, move reference material to L3. Effect-first — no splitting for aesthetics.
Automated structural health check across all skills with a single command. Catches issues before they ship.
27 auto-scannable, 26 require human judgment. Every rule traces to a real incident.
| Layer | Audit Target | Items | Auto | Coverage |
|---|---|---|---|---|
| L1 | Document Structure | 8 | 5 | |
| L2 | Architecture Consistency | 5 | 2 | |
| L3 | Code Quality | 8 | 6 | |
| L4 | Cross-File Consistency | 4 | 1 | |
| L5 | Doc-Code Alignment | 3 | 2 | |
| L6 | Dependency Integrity | 3 | 2 | |
| L7 | File Hygiene | 6 | 5 | |
| L8 | Security & Compliance | 3 | 2 | |
| L9 | Robustness & Degradation | 7 | 1 | |
| L10 | Content Quality | 6 | 1 | |
| Total | 53 | 27 |
See how mu-skill-creator compares to alternatives.
| Dimension | mu-skill-creator | Manual | Templates | AI Frameworks |
|---|---|---|---|---|
| Audit model | ✓ 53 items, 10 layers | ✗ None | ✗ None | Varies |
| Automated scanning | ✓ Shell script | ✗ | ✗ | Rare |
| Anti-pattern catalog | ✓ 34 items | ✗ | ✗ | Few |
| Context-aware loading | ✓ 3-layer | ✗ Single file | ✗ | Varies |
| Gated workflow | ✓ 8 stages | ✗ | ✗ | Rare |
| Stall detection | ✓ Quantitative | ✗ | ✗ | ✗ |
| Security scanning | ✓ Pre-publish | Manual only | ✗ | Varies |
| Trigger optimization | ✓ 10+10 test | ✗ | ✗ | ✗ |
| Open Source | ✓ MIT | — | — | Varies |
1. Clone the repository
git clone https://github.com/muippt/mu-skill-creator.git
2. Copy to your AI agent's skills directory
cp -r mu-skill-creator /path/to/your/agent/skills/
Tell your AI agent:
"Create a skill that does X" → full 8-stage creation workflow "Audit my-skill for quality issues" → 53-item 10-layer audit
Run the automated scanner:
bash scripts/skill-audit.sh <skill-name> # scan one skill bash scripts/skill-audit.sh # scan all skills bash scripts/skill-audit.sh --verbose # detailed output
Free, open source, and yours forever.