OPPO-PersonalAI/Agent_Foundation_Models — reverse-engineered prompt

I want this repo turned into a working, easy to run project so I can actually try the Chain of Agents idea without digging through research code. Please set it up so I can train or fine tune the Agent Foundation Model, download the provided web, multi hop QA, and code datasets, and run a simple example end to end. I also want a clear way to evaluate it on the included agent tasks, especially web interaction, question answering across multiple steps, and code execution.

Please make the setup practical on a normal research machine, fix anything broken, wire up the config paths, and add a small getting started flow with the exact commands I should run. If there are separate paths for supervised fine tuning and reinforcement learning, keep both available but make the easiest default path obvious. A simple inference demo would be great too, showing how one model can act like multiple cooperating agents. Look up current docs online if you need to.