ClawSec ClawHub Checker

Enhances the ClawSec suite's guarded skill installer with ClawHub reputation checks. Adds a second layer of security by checking VirusTotal Code Insight scores and other reputation signals before allowing skill installation.

Operational Notes

• - Required runtime: node, clawhub, INLINECODE2
• Depends on: installed INLINECODE3
• Side effects: setup_reputation_hook.mjs copies files into the installed suite and rewrites INLINECODE5
• Network behavior: reputation checks query ClawHub and may trigger remote metadata lookups during inspect/declined install flows
• Trust model: reputation scores are heuristic, not authoritative; keep the double-confirmation flow enabled

What It Does

1. 1. Wraps clawhub install - Intercepts skill installation requests
2. Checks VirusTotal reputation - Uses ClawHub's built-in VirusTotal Code Insight
3. Adds double confirmation - For suspicious skills (reputation score below threshold)
4. Integrates with advisory feed - Works alongside existing clawsec-suite advisories
5. Provides detailed reports - Shows why a skill is flagged as suspicious

Installation

This skill must be installed after clawsec-suite:

CODEBLOCK0

The setup script prints a preflight review before it mutates the installed suite files.

After setup, the checker adds enhanced_guarded_install.mjs and
guarded_skill_install_wrapper.mjs under clawsec-suite/scripts and updates the advisory
guardian hook. The original guarded_skill_install.mjs is not replaced.

Review the printed preflight summary before running setup. The script intentionally modifies the installed suite in place rather than operating on a temporary copy.

How It Works

Enhanced Guarded Installer

After setup, run the wrapper (drop-in path) or the enhanced script directly:
CODEBLOCK1

The enhanced flow:

1. 1. Advisory check (existing) - Checks clawsec advisory feed
2. Reputation check (new) - Queries ClawHub for VirusTotal scores
3. Risk assessment - Combines advisory + reputation signals
4. Double confirmation - If risky, requires explicit INLINECODE14

Reputation Signals Checked

1. 1. VirusTotal Code Insight - Malicious code patterns, external dependencies (Docker usage, network calls, eval usage, crypto keys)
2. Skill age & updates - New skills vs established ones
3. Author reputation - Other skills by same author
4. Download statistics - Popularity signals

Exit Codes

• - 0 - Safe to install (no advisories, good reputation)
• INLINECODE16 - Advisory match found (existing behavior)
• INLINECODE17 - Reputation warning (new - requires --confirm-reputation)
• INLINECODE19 - Error

Configuration

Environment variables:

• - CLAWHUB_REPUTATION_THRESHOLD - Minimum reputation score (0-100, default: 70)

Integration with Existing Suite

The checker enhances but doesn't replace existing security:

• - Advisory feed still primary - Known malicious skills blocked first
• Reputation is secondary - Unknown/suspicious skills get extra scrutiny
• Double confirmation preserved - Both layers require explicit user approval

Example Usage

CODEBLOCK2

Safety Notes

• - This is a defense-in-depth layer, not a replacement for advisory feeds
• VirusTotal scores are heuristic, not definitive
• False positives possible - Legitimate skills with novel patterns might be flagged
• Always review skill code before installing with INLINECODE21

Current Limitations

Missing OpenClaw Internal Check Data

1. 1. VirusTotal Code Insight - ✅ Our checker catches these flags
2. OpenClaw internal check - ❌ Not exposed via API (only on website)

Example from clawsec-suite page:

• - VirusTotal: "Benign" ✓
• OpenClaw internal check: "The package is internally consistent with a feed-monitoring / advisory-guardian purpose, but a few operational details and optional bypasses deserve attention before installing."

Our checker cannot access OpenClaw internal check warnings as they're not exposed via clawhub CLI or API.

Recommendation for ClawHub

• - clawhub inspect --json endpoint
• Additional API field for security tools
• Or include in clawhub install warning output

Workaround

Development

To modify the reputation checking logic, edit:

• - scripts/enhanced_guarded_install.mjs - Main enhanced installer
• INLINECODE27 - Reputation checking logic
• INLINECODE28 - Hook integration

License

GNU AGPL v3.0 or later - Part of the ClawSec security suite