senseipri/ShadowProtect — reverse-engineered prompt

Build me ShadowProtect, a self hosted security tool for teams running AI agents. I want it to work like Wireshark for agents, where I can wrap an existing agent with one simple Python call and then see what the agents say, what tools they call, errors, and suspicious behavior in a live web dashboard.

Please build the Python SDK, the backend API, and the frontend dashboard. The SDK should work with normal Python callables and common agent style objects, send events like agent start, messages, tool calls, and errors to the backend, and not require people to rewrite their agents.

The backend should inspect those events for prompt injection, hidden or encoded attacks, tool abuse, data leakage, collusion between agents, behavior drift, and tainted messages. It should also actively block or quarantine dangerous actions instead of only logging them.

Make it runnable locally with Docker compose, with a clean dashboard at localhost showing agents, live events, alerts, severity, and blocked actions. Look up current docs online if needed.