Dependency Guard 🛡️

Bitwarden CLI was compromised via npm. Your agent could be next. Scan before you install.

😱 The Problem

Your AI agent runs npm install some-mcp-server and 30 seconds later, your API keys are in someone else's database. This isn't hypothetical — it happened to Bitwarden CLI in April 2026.

Stat	Number
Packages with known vulnerabilities	86% of repos
Average detection time for compromised package	209 days
npm packages added daily	~1,200 (many are typosquats)
Agent-related supply chain incidents (2026)	47 and counting

AI agents are the perfect target: they install packages automatically, run with high privileges, and handle sensitive data. One bad npm install can compromise everything.

✅ The Solution

Dependency Guard scans npm/pip packages BEFORE installation. No dependencies. No build step. Copy the SKILL.md, and your agent will:

1. 🔍 Check metadata — Is this package legitimate or a typosquat?
2. ⚠️ Detect post-install scripts — The #1 attack vector for supply chain attacks
3. 🧬 Analyze dependency graph — What's hiding in the transitive chain?
4. 🏷️ Score risk — Quantified 0-100 risk score, not vibes
5. 🚑 Emergency response — What to do when you've already installed a bad package

🚀 Quick Start

# 30-second safety check before ANY install
npm view <package> scripts        # Check for postinstall hooks
npm view <package> author         # Verify the author
npm view <package> --json | jq '.time.created'  # New = suspicious
npm audit                         # Scan for known CVEs

If the package has postinstall scripts and you don't recognize the author — DO NOT INSTALL.

📋 What It Catches

Attack Type	How We Detect It	Real Example
Typosquatting	Edit distance from top-1000 packages	lodassh → lodash
Install-time execution	Scan for postinstall/preinstall scripts	Bitwarden CLI compromise
Dependency confusion	Check scoped vs unscoped registries	company-utils public vs private
Dormant malware	Sudden version spikes, new maintainer	ua-parser-js incident
Transitive attacks	Full dependency tree audit	event-stream via flatmap-stream
Obfuscated code	Scan for eval/atob/encoded strings	Credential stealers

🎯 Risk Score

Start at 100, subtract for risk factors:

  -30  Critical CVE found
  -20  Post-install scripts present
  -20  Typosquat match detected
  -15  Package < 30 days old
  -10  Single maintainer
   -5  Per suspicious transitive dependency

≥ 70  ✅ Safe to install
40-69 ⚠️  Install with caution
< 40  🚫 Do not install

🛠️ CI/CD Integration

# GitHub Actions - auto-scan new dependencies
- name: Dependency Guard
  run: |
    git diff origin/main -- package.json | grep '+    "' | while read line; do
      pkg=$(echo "$line" | sed 's/.*"\([^"]*\)".*/\1/')
      echo "Scanning: $pkg"
      npm view "$pkg" scripts --json
    done
    npm audit --production

📦 Zero Dependencies

This is a pure markdown skill. No npm. No pip. No builds. Your agent reads the SKILL.md and follows the procedures using tools already on your system (npm, pip, jq, grep).

Copy → paste → protected.

🔗 Related Skills

• skill-mcp-security-audit — Audit MCP server code for security issues
• skill-git-secret-sweep — Scan repos for leaked secrets and API keys

📰 Why Now

Supply chain attacks on AI agents are accelerating. In the past 30 days:

• Bitwarden CLI compromised via npm typosquatting (April 2026)
• 47 agent-related supply chain incidents reported
• Infisical launched Agent Vault — credential proxy for agents, showing the market demand

Don't wait until your agent's credentials are stolen. Start scanning today.

License

MIT — Use freely, scan everything, trust nothing.