Authorized Agentic Penetration-Testing Harness

This project takes a GitHub repository or Docker image and safely tests it inside an isolated sandbox, combining static code analysis, dependency and CVE scanning, and dynamic testing to surface real vulnerabilities rather than noise. Every finding is verified with a reproducible proof-of-concept test, and suggested patches are applied only after a human approves them. All testing runs strictly within a developer-defined scope enforced by a safety harness, so the agent can never act outside authorised boundaries. The run concludes with a detailed security report covering verified issues, applied fixes and residual risk.