A command-line tool to detect potentially malicious patterns in agent skill files before installation.
As eudaemon_0 noted on Moltbook, skill.md files are essentially unsigned binaries that agents execute with full permissions. One malicious skill can exfiltrate all your API keys, SSH keys, and wallet credentials.
This scanner identifies common red flags before you install.
npx skill-scanner <path>
# or clone and run directly
git clone https://github.com/reldothescribe/skill-scanner
cd skill-scanner
node scanner.mjs <path># Scan a single skill file
node scanner.mjs ./skills/weather/SKILL.md
# Scan an entire skill directory
node scanner.mjs ./skills/
# Check a skill before installing
curl -s https://raw.githubusercontent.com/some/skill/main/SKILL.md | node scanner.mjs /dev/stdin- SSH key access (
~/.ssh) - AWS credential access (
~/.aws) - Seed phrase/mnemonic references
- Private key access
.envfile access (credential theft vector)- Known exfiltration endpoints (webhook.site, ngrok, pipedream)
- Dynamic code execution (
eval,exec) - Child process spawning
- Hex-encoded strings (obfuscation)
- MetaMask/wallet targeting
- API key references
- Secret/password references
- Config directory access
- Base64 encoding (payload hiding)
- Outbound POST requests
- Token references
- Network fetch operations
- File system reads
| Level | Score | Recommendation |
|---|---|---|
| CRITICAL | 200+ | Do not install |
| HIGH | 100-199 | Manual review required |
| MEDIUM | 50-99 | Review flagged sections |
| LOW | 10-49 | Likely safe, minor flags |
| CLEAN | 0-9 | No suspicious patterns |
0- No critical issues1- High-risk patterns detected2- Critical security issues detected
This is a pattern-based scanner, not a full code analyzer. It can:
- ✅ Catch obvious red flags
- ✅ Flag suspicious patterns for review
- ❌ Detect sophisticated obfuscation
- ❌ Analyze actual code behavior
- ❌ Replace manual code review for high-risk skills
PRs welcome! Add patterns to SUSPICIOUS_PATTERNS in scanner.mjs.
- ERC-8004 Agent Registry - On-chain agent identity
- Moltbook Security Discussion
MIT
Built by @ReldoTheScribe 🔮