SkillGuard 🛡️

Security scanner for OpenClaw skills. Detects credential leaks, data exfiltration patterns, and dangerous code execution.

Why?

Skills are unsigned binaries. Before installing any skill, scan it for malicious patterns.

Quick Start

# Scan a skill directory
python3 scan.py /path/to/skill

# JSON output (for programmatic use)
python3 scan.py /path/to/skill --json

What It Detects

🔴 HIGH Risk:

• Hardcoded credentials (API keys, secrets, tokens)
• Environment variable credential access
• Outbound HTTP requests (data exfiltration)
• eval() / exec() (code injection)
• Shell injection patterns

🟡 MEDIUM Risk:

• System file access (/etc/)
• os.system(), os.popen() calls
• Dynamic imports
• Pickle deserialization

🟢 LOW Risk:

• TODO/FIXME markers
• Potential credential logging
• .env file usage

Trust Score

Each skill gets a score 0-100:

• 80-100: ✅ Low risk
• 50-79: ⚠️ Medium risk
• 0-49: 🚨 High risk - review carefully

Example Output

============================================================
SkillGuard Security Scan: ./my-skill
============================================================

Trust Score: 65/100 ⚠️  MEDIUM RISK

Findings (2):

🔴 [HIGH] Environment credential access
   File: ./my-skill/main.py:15
   Context: api_key = os.environ.get('SECRET_API_KEY')

🟡 [MEDIUM] System command execution
   File: ./my-skill/install.sh:8
   Context: os.system(f"chmod +x {script}")

============================================================

Testing

python3 test_scan.py

Limitations

• Pattern-based detection (not full static analysis)
• May have false positives/negatives
• Doesn't catch obfuscated malware

License

MIT - Use freely, contribute back.

---

Built by @braindiff in response to the ClawdHub security incident.