claw-shield

claw-shield is a production-ready OpenClaw security scanner for skills downloaded from ClawHub or other untrusted sources. It parses JavaScript with Babel, applies an extensible pattern registry, analyzes SKILL.md, and produces a terminal safety report with a bounded risk score.

Install

npm install

Usage

npx claw-shield scan ./path-to-skill

Optional JSON output:

npx claw-shield scan ./path-to-skill --json

Risk Model

• critical: +40 points
• high: +20 points
• medium: +10 points
• low: +5 points

Scores cap at 100 and map to:

• SAFE: 0-10
• SUSPICIOUS: 11-49
• DANGEROUS: 50-79
• CRITICAL: 80-100

Recommendations map to:

• SAFE: INSTALL
• SUSPICIOUS: INSTALL WITH CAUTION
• DANGEROUS or CRITICAL: DO NOT INSTALL

Coverage

The scanner detects:

• Dynamic eval, new Function, base64 staging, executable string assembly, and escape-heavy strings
• External fetch, axios, http, https, XMLHttpRequest, WebSocket, and DNS activity
• Environment harvesting plus env-to-network flows
• child_process, vm, dynamic require(), and suspicious filesystem access
• Prototype pollution
• SKILL.md permission mismatches, suspicious metadata requirements, and homepage/domain mismatch

Tests

npm test

The Jest suite covers malicious, clean, and suspicious fixtures, targeted rule validation, report formatting, and false-positive guards against popular npm packages.