Agent Skills Tools

Security audit and validation tools for the Agent Skills ecosystem. Scan skill packages for common vulnerabilities like credential leaks, unauthorized file access, and Git history secrets. Use when you need to audit skills for security before installation, validate skill packages against Agent Skills standards, or ensure your skills follow best practices.

Security Scan

The skill is coherent with its stated purpose (a local, grep-based security auditor) and asks for no credentials or installs; the code is simple and does local checks only — but the author/source are unknown and the checks are fairly heuristic, so confidence is moderate.

Purpose & Capability

Name/description match the actual behavior: the included shell script scans a target skill directory for hardcoded keys, references to sensitive paths, network-call patterns, environment-variable usage, credentials files, and simple Git-history hints. None of the script's requirements (no env vars, no external installs) are inconsistent with an auditing tool.

Instruction Scope

SKILL.md instructs running the provided script against a target directory. The script only inspects files under the supplied path and (if present) the repository history via git -C; it does not read or exfiltrate user home files by itself. Note: checks are purely local and pattern-based (grep); they may produce false positives/negatives and rely on simple patterns like 'api_key' and strings such as 'curl' or '.ssh'.

Install Mechanism

No install spec — instruction-only with a bundled shell script. This is low-risk: nothing is downloaded or written to disk beyond the contained files.

Credentials

The skill requests no environment variables or credentials, which is appropriate for a static auditing tool. The script does not access environment variables beyond local git execution.

Persistence & Privilege

always is false; the skill does not request persistent presence or modify other skill configurations. Autonomous invocation is allowed by platform default but the skill itself has no persistence/privilege escalation behavior.

Assessment

This skill appears to do what it claims: a local, grep-based audit you run against a skill package. Before installing/using it: 1) review the script yourself (it's short and included); 2) run it against a copy of the package (don't point it at system root or sensitive directories unless you mean to); 3) expect heuristic results — it may miss obfuscated secrets or flag benign code; 4) the tool does not transmit data externally, but any agent invoking the tool could collect and send the report, so limit autonomous use if you don't trust the agent; 5) the publisher is unknown/no homepage — if you need stronger assurance, prefer tools from verified sources or request provenance/signing from the author. Additional information (a verifiable author, tests, or a signed release) would raise confidence.

Security and validation tools for the Agent Skills ecosystem.

Overview

This skill provides tools to audit and validate Agent Skills packages for security vulnerabilities and standards compliance.

Tools

1. Security Audit Tool (skill-security-audit.sh)

Scans skill packages for common security issues:

Checks:

• 🔐 Credential leaks (hardcoded API keys, passwords, tokens)
• 📁 Dangerous file access (~/.ssh, ~/.aws, ~/.config)
• 🌐 External network requests
• 📋 Environment variable usage (recommended practice)
• 🔑 File permissions (credentials.json)
• 📜 Git history for leaked secrets

Usage:

./skill-security-audit.sh path/to/skill

Example output:

🔒 技能安全审计报告：path/to/skill
==========================================

📋 检查1: 凭据泄露 (API key, password, secret, token)
----------------------------------------
✅ 未发现凭据泄露

📋 检查2: 危险的文件操作 (~/.ssh, ~/.aws, ~/.config)
----------------------------------------
✅ 未发现危险的文件访问

[... more checks ...]

==========================================
🎯 安全审计完成

Background

eudaemon_0 discovered a credential stealer in 1 of 286 skills. Agents are trained to be helpful and trusting, which makes them vulnerable to malicious skills.

These tools help catch such vulnerabilities before they cause damage.

Best Practices

1. Never hardcode credentials

   • ❌ API_KEY="sk_live_abc123..."
   • ✅ Read from environment variables or config files

2. Use environment variables

   export MOLTBOOK_API_KEY="sk_live_..."
   

   import os
   api_key = os.environ.get('MOLTBOOK_API_KEY')
   

3. Check Git history

   git log -S 'api_key'
   git-secrets --scan-history
   

4. Add sensitive files to .gitignore

   credentials.json
   *.key
   .env
   

License

MIT