ClawHub lists many OpenClaw skills—making discovery fast and enabling instant capability upgrades. That aggregate value is real.
But the biggest pain point today is “default‑open” distribution. A past event (“ClawHavoc”) showed how risky this can be: approximately 12% of skills contained trojans, including families like the AMOS stealer.
Why this hub matters
Discovery
Find Skills quickly across categories; reduce setup time and get to results sooner.
Interoperability
Markdown‑based recipes fit agent ecosystems and can be adapted for multiple runtimes.
Community
Popular tasks evolve quickly as builders share fixes and improvements.
The risk: default‑open distribution
- ClawHavoc showed coordinated malware distribution via Skills; audits found ~12% trojan presence.
- Families like AMOS (credential/file stealers) were embedded in scripts and one‑liners.
- Default‑open must be paired with aggressive, transparent auditing to be safe for end users.
Security Scan Reports
Add a “security status” board for each Skill. Display whether code was audited, and flag suspicious behaviors like base64‑encoded one‑liner installers.
Audited
Static review complete; no obfuscation or network exfiltration detected.
Suspicious One‑Liner
Installer contains base64 strings, curl|bash chains, or encoded payloads requiring deeper review.
Unknown
No audit yet; run in a sandbox with minimal permissions until verified.
Permission Tier Labels
Label required local permissions explicitly so users can make informed decisions.
Direct Source Code Links
Add “View Source Code” links that jump to the core execution files (e.g., index.ts) on GitHub. This accelerates audits for security teams and developers.
Link Pattern
Repository → core file (index.ts, main.py); deep‑link to the specific path for review.
Audit Tips
Check for encoded payloads, unsafe eval/exec, network exfiltration, and unscoped file writes.
Build the audit‑first layer
- Parse Skill manifests and installer scripts; flag obfuscation and risky chains.
- Require explicit approvals for high‑risk permissions; default to sandboxed execution.
- Publish audit results publicly; make “trust” a visible, ongoing metric.