OWASP FinBot: Guardrail & Detection Framework for Agentic AI CTF

OWASP FinBot is a security learning tool that lets people practice attacking AI agent systems in a safe environment. Right now it only teaches how to attack. There is no way to practice defending. This project adds the defending side. I will build a system that watches what an AI agent does at key points, checks if the action looks safe or dangerous, and blocks or flags it if needed. I will also add a scoring system that gives points to defenders when they successfully stop an attack. On top of this, I will create five hands-on challenges where learners can practice both attacking and defending the same scenario. One extra challenge will cover a situation where a tool the agent uses turns out to be malicious. The final output includes the detection and blocking system, the scoring system, five paired challenges, example config files, full documentation, and a complete test suite.