Shannon — AI Pentester

An autonomous white-box penetration testing skill for Claude Code that analyzes source code, maps attack surfaces, and executes real exploits across 50+ vulnerability types before they reach production.

Shannon — Autonomous AI Pentester

A Claude Code skill that wraps the Shannon white-box penetration testing framework. It analyzes your source code, maps the attack surface, and executes real exploits to prove vulnerabilities before they reach production.

For authorized security testing only. Only run Shannon against systems you own or have explicit written permission to test.

What This Skill Does

Core Capabilities

  • 96.15% Exploit Success Rate: 100/104 exploits on the XBOW security benchmark
  • 50+ Vulnerability Types: SQL injection, XSS, SSRF, broken auth, broken authorization, and more
  • 5 OWASP Categories: Injection, XSS, SSRF, Broken Authentication, Broken Authorization
  • Multi-Agent Pipeline: Static analysis → recon → 5 parallel vulnerability agents → exploitation → reporting
  • No Exploit, No Report: Only reports vulnerabilities it can prove with a working proof-of-concept
  • Full Scope Control: Limit categories, exclude paths, configure concurrent pipeline count

Installation

BASH
npx skills add unicodeveloper/shannon

Requirements: Docker, Git, and an Anthropic/Bedrock/Vertex API key.

Usage

BASH
/shannon http://localhost:3000 myapp

Shannon will:

  1. Clone and analyze your source code
  2. Map the attack surface
  3. Run 5 parallel vulnerability agents
  4. Attempt to exploit each finding
  5. Generate a report with only proven vulnerabilities

Authentication Support

Shannon handles complex auth flows:

BASH
/shannon http://localhost:3000 myapp --auth form
/shannon http://localhost:3000 myapp --auth sso
/shannon http://localhost:3000 myapp --auth totp

Bundled Tools

All containerized in Docker — no manual setup:

  • Nmap — Network and port scanning
  • Subfinder — Subdomain discovery
  • WhatWeb — Technology fingerprinting
  • Schemathesis — API fuzzing and schema testing
  • Chromium/Playwright — Browser-based exploit execution

What to Expect

MetricValue
Benchmark success rate96.15% (100/104)
Typical duration~1–1.5 hours per full pentest
Estimated cost~$50 with Claude Sonnet

Scope Control

Limit what Shannon tests:

BASH
# Test only injection vulnerabilities
/shannon http://localhost:3000 myapp --categories injection

# Exclude specific paths
/shannon http://localhost:3000 myapp --exclude /admin,/health

# Set concurrent pipeline count
/shannon http://localhost:3000 myapp --concurrency 3

Shannon is a white-box tester — it works best with source code access. For black-box testing, results may vary.