🛡️ clawhub-guard

Pre-install security scanner for ClawHub skills.
Scan before you install. Never trust blindly.

Why?

ClawHub hosts thousands of community skills — but not all are safe. In early 2026, the ClawHavoc campaign distributed 341 malicious skills through the registry, stealing credentials and installing malware.

clawhub-guard adds a mandatory security checkpoint before every install.

What It Does

$ clawhub-guard install summarize

📦 Installing summarize for pre-scan...
✓ Installed [email protected]

🔍 Running security scan on summarize...

═══════════════════════════════════════════════════════
  SECURITY SCAN REPORT — clawhub-guard
═══════════════════════════════════════════════════════
  Target:     ~/.openclaw/workspace/skills/summarize
  Score:      94/100
  Threshold:  70/100
  Verdict:    ✅ PASS
  Engines:    1/4 available
  Findings:   2
───────────────────────────────────────────────────────
    ⚪ low: 2
───────────────────────────────────────────────────────
  • [low] tool-shadowing: Redirects from another tool
  • [low] excessive-perms: Too many permissions declared
═══════════════════════════════════════════════════════

✅ Safe! summarize is ready to use.

If the score is too low:

❌ BLOCKED — Uninstalling malicious-skill for safety.
💡 Tip: Review the findings above. Use --force to install anyway.

Install

npm install -g clawhub-guard

Usage

# Install a ClawHub skill with automatic pre-scan
clawhub-guard install <skill-name>

# Scan an already-installed skill
clawhub-guard scan <skill-name>

# Scan a local skill directory
clawhub-guard scan --local ./my-skill/

# Custom risk threshold (default: 70)
clawhub-guard install <skill-name> --threshold 80

# Force install (skip security scan)
clawhub-guard install <skill-name> --force

# JSON output for automation
clawhub-guard scan <skill-name> --json

Scoring

| Score | Verdict | Action | |-------|---------|--------| | 90–100 | ✅ PASS | Safe to install | | 70–89 | ⚠️ WARN | Installed with warnings — review findings | | 40–69 | ❌ BLOCK | Automatically uninstalled — review report | | 0–39 | ❌ BLOCK | Automatically uninstalled — do not use |

Powered By

• AgentShield — 30 security rules covering credential theft, backdoors, prompt injection, obfuscation, and more.

Recommended Workflow

# 1. Always use clawhub-guard instead of raw openclaw skills install
clawhub-guard install <skill-name>

# 2. If blocked, review the scan report
clawhub-guard scan <skill-name> --json | less

# 3. Only force-install if you've manually reviewed the code
clawhub-guard install <skill-name> --force

Security

This tool is itself a security product. It:

• Runs locally only — no telemetry, no external calls beyond agent-shield
• Does not modify your OpenClaw config
• Does not read your credentials or session data
• Automatically uninstalls skills that fail the security threshold

License

MIT © taiwanape

"Trust is earned in milliseconds, lost in microseconds, and clawed back never." — clawhub-guard