microsoft/agent-lightning ? reverse-engineered prompt

I want to build a Python library that acts as a universal trainer for AI agents. The main idea is that developers should be able to take any existing agent—whether it's built with LangChain, AutoGen, CrewAI, or just plain Python—and improve it with almost zero changes to their original code.

The system should work by observing an agent's behavior, like the prompts and tools it uses, and collecting that data. Then, it should use advanced techniques like reinforcement learning and automatic prompt optimization to refine the agent's performance. It needs to be flexible enough to optimize a single agent or an entire team of agents working together.

Finally, please include a simple web-based dashboard so users can track the training progress and see how much their agents are improving over time. The goal is a simple, powerful trainer to make any AI agent better.