Multi-agent red team security scanner powered by Claude. Four AI agents collaborate to scan deliberately vulnerable web apps (DVWA, OWASP Juice Shop) and stream results live to a dark terminal-style UI.
| Agent | Role | Tools |
|---|---|---|
| Coordinator | Creates engagement plan | None (single API call) |
| Recon | Enumerates attack surface | fetch_headers, fetch_robots, check_common_paths, detect_tech_stack |
| Scanner | Runs OWASP Top 10 checks | test_sqli, test_xss, check_security_headers, test_idor, check_cors |
| Exploit Mapper | Maps findings to CVE/CWE | None (single API call) |
# 1. Start a target (pick one)
docker run -d -p 8080:80 vulnerables/web-dvwa
# or
docker run -d -p 3000:3000 bkimminich/juice-shop
# 2. Set your API key
export ANTHROPIC_API_KEY=sk-ant-...
# 3. Install and run
pip install -r requirements.txt
python3 main.pyOpen http://localhost:8888 — select your target, pick a mode, and launch.
- Full Engagement — all four agents in sequence
- Recon Only — coordinator + recon
- Scan Only — coordinator + scanner
- Exploit Map Only — exploit mapper on previous findings
- Python 3.11+, FastAPI, uvicorn
- Anthropic SDK (claude-sonnet-4-20250514)
- No LangChain/LangGraph — pure custom agent loop
- All HTTP probing via stdlib
urllib - SSE streaming to browser
This tool is designed exclusively for authorized security testing of local lab environments. Only use against systems you own or have explicit written permission to test.