ClawHub Security Check#

Mandatory pre-installation security assessment for ClawHub skills with optional sandbox testing.

When to Use#

• User wants to install a ClawHub skill ( openclaw skill install, npx clawhub install)
• User asks about skill safety or wants to check a skill before installing
• BEFORE executing any skill installation command

Workflow#

1. Extract Skill Info#

Parse the skill identifier:

• Full URL: https://clawhub.ai/author/skill-name
• Short: skill-name or author/skill-name

2. Fetch ClawHub & GitHub Data#

bash

web_fetch("https://clawhub.ai/{author}/{skill-name}")
web_fetch("https://github.com/{author}/{repo}")

Extract: author, description, download count, stars, last commit date, verified status, file structure.

3. Calculate Risk Score#

See references/risk-scoring.md for the full scoring table.

Score is 0–100 (higher = safer):

• 80–100: Very safe → skip sandbox, approve
• 60–79: Safe → approve, optional sandbox
• 40–59: Moderate → sandbox required
• 20–39: High risk → sandbox required
• 0–19: Very high risk → reject immediately

4. Sandbox Testing (Conditional)#

Run sandbox IF score is 20–79 AND skill has executables or unknown author.

Skip IF score ≥ 80 (trusted), documentation-only, or score < 20 (reject).

Spawn an isolated sub-agent session to install and analyze the skill:

• List all files and executables
• Grep for suspicious patterns ( curl|sh, eval, exec, rm -rf, sudo, ~/.ssh, ~/.aws, base64.*decode)
• Check for network calls
• Return structured JSON with findings

See references/sandbox-procedure.md for the full sub-agent task template.

5. Generate Report#

See references/report-templates.md for all report formats.

Present a concise security report with verdict (SAFE / NEEDS REVIEW / NOT SAFE), author info, trust rating, and a clear action instruction. If user asks "why" or "details", show expanded findings.

6. Handle User Response#

• yes → run openclaw skill install [skill-name]
• no → cancel, suggest alternatives
• details → show full analysis, then re-ask

Safety Rules#

• Never install without explicit user confirmation
• Security check cannot be skipped or bypassed
• Err on caution for unknown sources
• If no internet, block installation entirely
• Limit concurrent sandbox sessions to 1