komnsensei/freebrain ? reverse-engineered prompt

Build me a Python project that acts like a self rewriting agent and can run a long drift study over many cycles. I want it to keep a clear record of what it changed, what evidence it used, and whether each run stayed stable or drifted over time. It should have a main runtime, an autonomy loop, guardrails so it does not wander off, and simple metrics and audit logs that help me understand how the system is behaving.

Please make it easy to run tests, inspect the agent’s state, and evaluate the retrieval or memory pieces if they are part of the flow. If something depends on current best practices or recent docs, look those up online. Keep the code clean, practical, and focused on being able to swap compute or runtime pieces later without changing the core logic.