opena2a-org/damn-vulnerable-ai-agent ? reverse-engineered prompt

Build me a local demo for an intentionally vulnerable AI agent playground that I can run with Docker. I want a dashboard on port 9000 that shows a fleet of agents, their security levels, and live attack stats, plus an attack lab where I can walk through different security scenarios and see what happens step by step.

Make it feel like a training tool for red teaming AI agents, with multiple agents that behave differently, some weak, some hardened, and some protected by capability controls. It should support testing common issues like prompt injection, data leaks, tool abuse, memory problems, and multi agent trust bugs. If it helps, look up any current docs online so the setup matches the real project.

I also want easy ways to run attacks against the agents directly from the command line and to see the results in the dashboard. Keep it clearly labeled as insecure and meant only for training, not production.