ClawHub Skill Vetting: Trust but Verify
ClawHub is OpenClaw’s community skill marketplace — think of it like an app store for agent capabilities. There are thousands of skills available: email management, calendar integration, research tools, home automation, and more.
The problem: over 400 malicious or unsafe skills have been identified in ClawHub. These range from skills that exfiltrate API keys to ones that inject hidden instructions into the agent’s context.
How to Vet a Skill Before Installing
1. Check the Author
- Is the author a known contributor to the OpenClaw project?
- Do they have other well-reviewed skills?
- Is their GitHub profile real and active?
2. Check the Stars and Downloads
- Skills with very few downloads could be new and unvetted
- High stars with low recent downloads could indicate a stale or abandoned skill
- Look for a consistent download trend, not just a spike
3. Read the Code
This is the most important step. Before installing any skill:
# Clone the skill repo first — do NOT install blindly
git clone https://github.com/author/skill-name
cd skill-name
# Look at what files exist
find . -type f -name "*.ts" -o -name "*.js" -o -name "*.py"
# Search for suspicious patterns
grep -r "fetch\|axios\|http\|request" --include="*.ts" --include="*.js"
grep -r "env\|secret\|key\|token\|password" --include="*.ts" --include="*.js"
grep -r "eval\|exec\|spawn\|child_process" --include="*.ts" --include="*.js"Red flags to look for:
| Pattern | Why It Is Suspicious |
|---|---|
| Outbound HTTP requests to unknown domains | Could be exfiltrating data |
| Reading environment variables | Could be stealing API keys |
Using eval() or exec() | Could execute arbitrary code |
| Obfuscated or minified source code | Hiding malicious behavior |
| Requesting permissions beyond stated purpose | A “weather” skill should not need shell access |
No README or documentation | Low effort, potentially malicious |
4. Check Recent Activity
# Look at recent commits
git log --oneline -20
# Check if there were any suspicious recent changes
git diff HEAD~5..HEADA skill that was dormant for months and suddenly received a large update could have been compromised (supply chain attack).
5. Use the Community
The OpenClaw Discord and GitHub Discussions are active communities. Before installing a skill you are unsure about, search for it in these forums. Others may have already reviewed it.
Installing Skills Safely
Even after vetting, install skills with restricted permissions:
# Install with restricted permissions
openclaw skill install author/skill-name --sandbox
# Review what permissions the skill requests
openclaw skill inspect author/skill-nameThe --sandbox flag runs the skill in an isolated environment where it cannot access your file system, environment variables, or other skills.
Next Steps
With your skills vetted, make sure the rest of your setup is hardened:
- Network Security & Checklist — Lock down your gateway and work through the full security checklist
- DM Policies & Pairing — Control who can talk to your agent
- Sandboxing — Contain the blast radius with sandbox mode and Docker
- API Keys & Tool Policies — Secure your credentials and control tool permissions