OpenClaw Skills & ClawHub Safety

ClawHub is OpenClaw’s skill marketplace with 5,700+ skills. But security researchers have found that approximately 341 skills distribute malware — including credential theft, crypto mining, and reverse shell payloads. This is known as the ClawHavoc campaign.

Curate-Me provides a ClawHavoc-hardened skill scanner that grades every skill before installation.

The ClawHub Security Problem

Threat	Description	Impact
Credential theft	Skills that exfiltrate API keys, SSH keys, and tokens	Account compromise
Crypto mining	Hidden mining scripts consuming your VPS resources	$100s/mo in compute waste
Reverse shells	Backdoor access to your execution environment	Full system compromise
Data exfiltration	Skills that send conversation data to external servers	Privacy violation
Supply chain attacks	Popular skills with malicious updates	Wide-blast-radius compromise

How Our Scanner Works

Every skill import goes through a 3-layer security pipeline:

Layer 1: Static Pattern Analysis

YARA-style rules scan for:

Shell injection patterns (os.system, subprocess, eval, exec)
Crypto wallet addresses and mining pool connections
Outbound connection attempts to suspicious domains
Base64-encoded payloads (common obfuscation technique)
File system access to sensitive paths (.env, .ssh, .aws)

Layer 2: Dependency Audit

Cross-reference all dependencies against a known-malicious blocklist
Check for typosquatted package names
Verify dependency version pins (unpinned deps = supply chain risk)
Flag dependencies with known CVEs

Layer 3: VirusTotal Integration

Submit skill content hash to VirusTotal
Check against 70+ antivirus engines
Flag any detection as CAUTION or BLOCKED

Security Verdicts

Every scanned skill receives a verdict:

Verdict	Badge	Meaning
SAFE	Green	No findings. Clear to install.
CAUTION	Amber	Non-critical findings. Review recommended. Admin can approve.
BLOCKED	Red	Critical findings. Cannot be installed.
UNSCANNED	Gray	Not yet scanned. Treated as untrusted.

Verified Skills

Curate-Me maintains a curated allowlist of 17 verified skills that have passed comprehensive security review:

Skill	Category	Description
`github`	Developer	GitHub integration (PRs, issues, code review)
`slack`	Communication	Slack messaging and channel management
`google-drive`	Productivity	Google Drive file management
`web-search`	Search	Web search via multiple providers
`puppeteer`	Automation	Browser automation and scraping
`docker`	DevOps	Docker container management
`postgres`	Database	PostgreSQL query execution
`redis`	Database	Redis key-value operations

Only verified skills can be installed on managed runners without admin approval.

Using the Skill Gallery

The Skill Gallery in the dashboard (/runners/gallery) provides:

Browse — Search and filter skills by category, verdict, and popularity
Scan — View detailed security findings for any skill
Import — One-click import with automatic security scan
Approve/Reject — Admin workflow for CAUTION-rated skills
Rescan — Re-run security checks with the latest scan engine
Stats — Aggregate security statistics across all imports

Free Skill Scanner

Scan any ClawHub skill URL for free at curate-me.ai/scanner . No account required. Results in under 10 seconds.

Best Practices

Only install verified skills on production runners
Review CAUTION findings before approving — understand what each finding means
Never install BLOCKED skills — even if they look useful
Rescan periodically — new threat patterns are added to the scanner weekly
Use model allowlists alongside skill scanning — defense in depth
Monitor skill behavior — check the audit trail for unexpected network calls