OpenClaw Skill

skulk-skill-scanner

Scan OpenClaw skill folders for security red flags before installing or publishing. Detects data exfiltration, credential theft, prompt injection, destructive commands, obfuscation, privilege escalation, and supply chain risks. Use when: evaluating a skill from ClawHub before install, auditing your own skills before publishing, or reviewing any SKILL.md for safety. NOT for: general code review or vulnerability scanning of non-skill codebases.

Install

$npx clawhub@latest install skulk-skill-scanner
View on GitHubv1.0.1
All-time installs5
Active installs5
Stars0

Skill Scanner

Security scanner for OpenClaw agent skills. Static analysis for red flags.

Usage

bash
node scripts/scanner.js <path-to-skill> [--verbose] [--json] [--summary] [--ignore <path>] [--include-self]

Examples

bash
# Scan a downloaded skill folder before enabling it
clawhub inspect some-skill
node scripts/scanner.js ./skills/some-skill --verbose

# Scan your own skill before publishing
node scripts/scanner.js ./skills/my-skill

# JSON output for automation
node scripts/scanner.js ./skills/my-skill --json

# One-line summary output for heartbeat checks
node scripts/scanner.js ./skills/my-skill --summary

# Include scanner internals (off by default to reduce self-scan noise)
node scripts/scanner.js ./skills/skulk-skill-scanner --include-self

What It Catches

SeverityFlags
🔴 CriticalData exfiltration, credential access, safety overrides, destructive commands
🟠 HighObfuscation (base64/eval), unknown network access, env scanning, privilege escalation, hidden instructions
🟡 MediumWrites outside workspace, package installs (supply chain), messaging on user's behalf, persistent timers/automation
🔵 InfoAPI key references, broad tool access requests

Scoring

  • Each unique rule deducts points: critical=30, high=15, medium=5, info=0
  • Score 75-100: ✅ PASS
  • Score 50-74: ⚠️ WARN
  • Score 0-49 or any critical: ❌ FAIL
  • Exit code 1 on FAIL (CI-friendly)

Safe Domain Allowlist

Known legitimate API domains are allowlisted to reduce false positives on network-related rules. Edit the SAFE_DOMAINS array in scripts/scanner.js to customize.

Limitations

This is static pattern matching — it catches obvious and moderately obfuscated attacks but cannot detect:

  • Sophisticated multi-step social engineering
  • Runtime-generated URLs or dynamic exfiltration
  • Attacks that look identical to legitimate skill behavior

It's a first line of defense, not a guarantee. Always review skills manually before granting sensitive access.

Created by

@adainthelab
View on ClawHub

Persistent memory

Give your OpenClaw agent a memory layer

Mem0 remembers users and context across sessions so you send fewer tokens and get better answers.

Try Mem0Mem0 + OpenClaw guide