What Happens When You Red-Team Your Own Agent

I built Agent Containment Protocol to answer a question that kept bugging me: if you give an AI agent API access, how do you know it's staying within bounds? The concept is straightforward. Give an ag