lev-os Skill Decompile — SKILL.md Security Scanner
Converts freeform SKILL.md files into normalized, machine-auditable YAML structures. Bridges the gap between human-authored skill definitions and deterministic security scanning.
Usage
# Single file
python3 ~/.agents/skills-db/security/skill-decompile/decompile.py ~/.agents/skills/auth-sniffer/SKILL.md
# Single file, JSON output
python3 ~/.agents/skills-db/security/skill-decompile/decompile.py ~/.agents/skills/auth-sniffer/SKILL.md --output json --verbose
# Scan entire skills directory with summary report
python3 ~/.agents/skills-db/security/skill-decompile/decompile.py --dir ~/.agents/skills/ --report
# Full directory scan as YAML
python3 ~/.agents/skills-db/security/skill-decompile/decompile.py --dir ~/.agents/skills/ --output yaml
What It Extracts
| Category | Details |
|---|---|
| meta | name, description, skill_type, source (local/external) |
| permissions | allowed_tools, requests for bash/write/network/env |
| surfaces | URLs, shell commands, env vars, file paths |
| risk_signals | eval, base64, obfuscation, exfil patterns, prompt injection |
| complexity | line count, code blocks, reference files, decision tree depth |
| verdict | risk_score (0-100), risk_level, specific flags |
Risk Scoring
- 0-20: safe — no concerning patterns
- 21-40: low — minor signals, likely benign
- 41-60: medium — review recommended
- 61-80: high — manual audit required
- 81-100: critical — block until reviewed
When to Use
- Before installing a third-party skill
- Periodic security audit of skill inventory
- CI/CD gate for skill contributions
- Comparing skill versions for permission drift