clawsec-clawhub-checker

# 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.

## What It Does

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`:

```bash
# First install the suite
npx clawhub@latest install clawsec-suite

# Then install the checker
npx clawhub@latest install clawsec-clawhub-checker

# Run the setup script to integrate with clawsec-suite
node ~/.openclaw/skills/clawsec-clawhub-checker/scripts/setup_reputation_hook.mjs

# Restart OpenClaw gateway for changes to take effect
openclaw gateway restart
```

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.

## How It Works

### Enhanced Guarded Installer

After setup, run the wrapper (drop-in path) or the enhanced script directly:
```bash
# Recommended drop-in wrapper
node scripts/guarded_skill_install_wrapper.mjs --skill some-skill --version 1.0.0

# Or call the enhanced script directly
node scripts/enhanced_guarded_install.mjs --skill some-skill --version 1.0.0
```

The enhanced flow:
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 `--confirm-reputation`

### Reputation Signals Checked

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)
- `42` - Advisory match found (existing behavior)
- `43` - Reputation warning (new - requires `--confirm-reputation`)
- `1` - 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

```bash
# Try to install a skill
node scripts/guarded_skill_install_wrapper.mjs --skill suspicious-skill --version 1.0.0

# Output might show:
# WARNING: Skill "suspicious-skill" has low reputation score (45/100)
# - Flagged by VirusTotal Code Insight: crypto keys, external APIs, eval usage
# - Author has no other published skills
# - Skill is less than 7 days old
# 
# To install despite reputation warning, run:
# node scripts/guarded_skill_install_wrapper.mjs --skill suspicious-skill --version 1.0.0 --confirm-reputation

# Install with confirmation
node scripts/guarded_skill_install_wrapper.mjs --skill suspicious-skill --version 1.0.0 --confirm-reputation
```

## 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 `--confirm-reputation`

## Current Limitations

### Missing OpenClaw Internal Check Data
ClawHub shows two security badges on skill pages:
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
To enable complete reputation checking, ClawHub should expose internal check results via:
- `clawhub inspect --json` endpoint
- Additional API field for security tools
- Or include in `clawhub install` warning output

### Workaround
Our heuristic checks (skill age, author reputation, downloads, updates) provide similar risk assessment but miss specific operational warnings about bypasses, missing signatures, etc. Always check the ClawHub website for complete security assessment.

## Development

To modify the reputation checking logic, edit:
- `scripts/enhanced_guarded_install.mjs` - Main enhanced installer
- `scripts/check_clawhub_reputation.mjs` - Reputation checking logic
- `hooks/clawsec-advisory-guardian/lib/reputation.mjs` - Hook integration

## License

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