vbguard

The definitive security scanner for AI-generated code.

45% of AI-generated code ships with known vulnerabilities. The Moltbook breach. The Lovable auth bypass. Hardcoded Supabase keys. Hallucinated npm packages. These aren't hypothetical — they're happening right now.

vbguard catches what Snyk, Semgrep, and SonarQube can't: the security mistakes unique to AI coding tools (Cursor, Claude Code, Copilot, Lovable, Bolt, Replit, v0).

npx vbguard .

No config. No account. No API key. Runs in milliseconds.

What It Catches

12 Scanner Categories

| Scanner | What It Detects | Severity | |---------|----------------|----------| | Hallucinated Packages | AI-invented npm/PyPI packages that don't exist (slopsquatting), typosquats of popular packages | Critical | | Auth Flow Analyzer | Unprotected routes, inverted auth logic, JWT without expiration, weak JWT secrets, tokens in localStorage, WebSocket auth gaps | Critical-High | | Vibe-Code Patterns | Security TODOs left as placeholders, hardcoded admin credentials, commented-out security code, AI signature markers | Critical-High | | Input Validation | SQL injection, NoSQL injection, command injection, path traversal, XSS, ReDoS, file upload bypasses | Critical | | Hardcoded Secrets | API keys (OpenAI, Stripe, AWS, GitHub, etc.), DB connection strings, JWTs, private keys | Critical | | Frontend-Exposed Secrets | Server secrets in client-side code, Stripe secret keys in React components, service role keys in browser | Critical | | Dangerous Functions | eval() with user input, pickle.loads(), unsafe YAML, SQL via template literals | Critical | | Crypto Analyzer | MD5/SHA1 for passwords, Math.random() for tokens, hardcoded encryption keys, ECB mode, disabled SSL | Critical-High | | Next.js Scanner | API keys in "use client", NEXT_PUBLIC_ on secrets, Server Actions without validation, SSR data leaks | Critical-High | | Supabase Scanner | Service role key in client code, missing RLS, unfiltered select('*'), signup without email verification | Critical-High | | Firebase Scanner | Permissive Firestore/RTDB/Storage rules, Admin SDK in client code, Cloud Functions without auth | Critical | | Docker Scanner | Running as root, privileged containers, exposed DB ports, .env in images, no healthchecks | Medium-Critical |

Plus

• Dangerous Dependencies — Compromised packages (event-stream), deprecated libs AI still suggests
• Missing .gitignore — .env files about to be committed, secrets in git history
• Permissive Configs — cors(*), debug=True, open database rules
• Missing Security Middleware — No rate limiting, no helmet, no auth middleware

Comparison

| Feature | vbguard | Snyk | Semgrep | GitGuardian | TruffleHog | |---------|---------|------|---------|-------------|------------| | Hallucinated package detection | Yes | No | No | No | No | | Auth flow analysis | Yes | No | Partial | No | No | | AI-specific patterns (TODOs, placeholders) | Yes | No | No | No | No | | Broken access control detection | Yes | No | Partial | No | No | | Hardcoded secrets | Yes | Partial | Partial | Yes | Yes | | Supabase RLS checks | Yes | No | No | No | No | | Firebase rules analysis | Yes | No | Partial | No | No | | Next.js-specific checks | Yes | No | Partial | No | No | | JWT security analysis | Yes | No | Partial | No | No | | Zero config | Yes | No | No | No | Yes | | Offline mode | Yes | No | Yes | No | Yes | | Free | Yes | Freemium | Freemium | Freemium | Yes |

Framework Support

• Next.js — Server Actions, API routes, "use client" secrets, NEXT_PUBLIC_ misuse
• Supabase — RLS policies, service role keys, anon key exposure, unfiltered queries
• Firebase — Firestore/RTDB/Storage rules, Admin SDK, Cloud Functions auth
• Express — Route auth, rate limiting, CORS, helmet, middleware chain
• FastAPI — Dependency injection auth, debug mode, CORS configuration
• Flask/Django — Secret keys, debug mode, CSRF protection

CLI Reference

# Basic scan
npx vbguard .

# Windows
npm exec vbguard .

# Only critical issues
vbguard . --severity=critical

# JSON output (for scripts)
vbguard . --json

# SARIF output (for GitHub Security tab)
vbguard . --ci

# HTML report
vbguard . --format=html

# Markdown report
vbguard . --format=md

# Security score
vbguard . --score

# Score badge URL
vbguard . --badge

# Generate fix suggestions file
vbguard . --fix

# Only scan changed files (git diff)
vbguard . --diff

# Watch mode (re-scan on file change)
vbguard . --watch

# Skip network checks (offline mode)
vbguard . --offline

# Ignore specific rules
vbguard . --ignore=secret/generic-jwt-token,vibe/ai-generated-marker

# Hide fix suggestions
vbguard . --no-fix

All Flags

| Flag | Description | |------|-------------| | --json | Output JSON | | --format=X | Output format: table (default), json, sarif, md, html | | --severity=X | Minimum severity: low, medium, high, critical | | --ci | SARIF output for CI/CD | | --score | Show security score (0-100) | | --badge | Output shields.io badge URL | | --fix | Generate .vbguard-fixes.md with fix suggestions | | --diff | Only scan files changed since last commit | | --watch | Watch mode with auto-rescan | | --offline | Skip network checks | | --skip-network | Alias for --offline | | --ignore=X | Comma-separated rules/paths to ignore | | --no-fix | Hide fix suggestions in output | | -h, --help | Show help | | -v, --version | Show version |

CI/CD Integration

GitHub Actions (Reusable Action)

name: Security Scan
on: [push, pull_request]

permissions:
  security-events: write
  pull-requests: write

jobs:
  vbguard:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: YOUR_USERNAME/vbguard@v1
        with:
          severity: low
          fail-on: high
          upload-sarif: true
          comment-on-pr: true

GitHub Actions (Simple)

name: Security Scan
on: [push, pull_request]

jobs:
  vbguard:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
      - run: npx vbguard . --severity=high

Pre-commit Hook (Husky)

# .husky/pre-commit
npx vbguard -- --diff --severity=critical --offline

Pre-commit Framework

# .pre-commit-config.yaml
repos:
  - repo: https://github.com/YOUR_USERNAME/vbguard
    rev: v1.0.0
    hooks:
      - id: vbguard

Configuration

.vbguardrc.json

{
  "severity": "medium",
  "ignore": ["secret/example-api-key"],
  "ignorePaths": ["test/", "scripts/", "*.test.js"],
  "format": "table",
  "enableNetworkChecks": true
}

vbguard.config.js

module.exports = {
  severity: 'medium',
  ignore: ['vibe/ai-generated-marker'],
  ignorePaths: ['test/', 'scripts/'],
  format: 'table',
  enableNetworkChecks: true,
};

.vbguardignore

Works like .gitignore — one pattern per line:

# Ignore test files
test/
*.test.js
*.spec.js

# Ignore specific rules
secret/generic-jwt-token

# Ignore scripts
scripts/

Inline Ignores

// vbguard-ignore-next-line
const hash = crypto.createHash("md5").update(data).digest("hex");

Security Score

--score outputs a single number from 0-100:

| Score | Rating | Color | |-------|--------|-------| | 90-100 | Secure | Green | | 70-89 | Needs work | Yellow | | 50-69 | At risk | Orange | | 0-49 | Critical | Red |

Scoring: CRITICAL = -15pts, HIGH = -8pts, MEDIUM = -3pts, LOW = -1pt.

Use --badge to get a shields.io URL for your README.

How It Works

vbguard uses pattern matching (regex + structural analysis) against a curated ruleset of AI-specific vulnerability patterns. No AI, no API calls, no data leaves your machine.

The hallucinated package detector optionally queries npm and PyPI registries to verify packages exist. Use --offline to skip.

The ruleset is based on real-world breaches and research:

• DryRun Security's March 2026 report (broken access control in every AI-generated app)
• The Moltbook breach (Supabase RLS misconfiguration)
• Tenzai's 2025 study (69 vulnerabilities across 5 AI coding tools)
• Escape.tech's scan of 5,600 vibe-coded apps
• Georgia Tech's Vibe Security Radar

Scanner Architecture

src/scanners/
  secrets.js              # Hardcoded API keys, tokens, connection strings
  dangerous-defaults.js   # Missing auth, rate limiting, CORS, headers
  dangerous-functions.js  # eval, pickle, SQL injection, XSS
  exposed-frontend.js     # Server secrets in client-side code
  permissive-configs.js   # Supabase, Firebase, Docker misconfigs
  dependencies.js         # Compromised/deprecated packages
  gitignore.js            # Missing .gitignore entries
  hallucinated-packages.js # AI-hallucinated packages, typosquats
  auth-flow.js            # Auth logic, JWT, OAuth, WebSocket auth
  vibe-patterns.js        # AI-specific code patterns
  input-validation.js     # SQL/NoSQL/command injection, XSS, ReDoS
  nextjs.js               # Next.js-specific security
  supabase.js             # Supabase-specific security
  firebase.js             # Firebase-specific security
  docker.js               # Docker/compose security
  crypto.js               # Cryptographic security

Contributing

Contributions welcome. If you've found a vulnerability pattern that AI tools commonly introduce, open a PR to add it.

Each scanner is a standalone module that exports a function receiving a file context object:

function scanMyPattern(ctx) {
  const { content, relativePath, ext, basename } = ctx;
  const findings = [];
  // ... pattern matching ...
  findings.push({
    rule: 'category/rule-name',
    severity: 'critical', // critical | high | medium | low
    file: relativePath,
    line: lineNumber,
    message: 'What is wrong and why it matters.',
    fix: 'How to fix it.',
  });
  return findings;
}

Changelog

v1.0.0

• Hallucinated package detection (npm + PyPI registry checks, typosquat detection)
• Auth flow analyzer (12+ patterns: inverted logic, JWT issues, WebSocket auth, OAuth)
• Vibe-code pattern detector (security TODOs, placeholder data, AI markers, error leaks)
• Input validation scanner (SQL/NoSQL/command injection, path traversal, XSS, ReDoS)
• Framework scanners: Next.js, Supabase, Firebase, Docker (enhanced)
• Crypto analyzer (weak hashing, Math.random, hardcoded keys, ECB, SSL, cookies)
• Smart CLI: --fix, --diff, --watch, --ci (SARIF), --score, --badge, --format
• Pre-commit hook support (Husky + pre-commit framework)
• GitHub Action (reusable action with SARIF upload + PR comments)
• HTML/Markdown/SARIF report output
• .vbguardrc.json + vbguard.config.js configuration
• .vbguardignore + inline ignores
• Scan comparison (new/fixed since last scan)
• 130+ tests

v0.4.0

• Initial release with secrets, dangerous functions, configs, dependencies

License

MIT