AgentR1/Agent-R1 ? reverse-engineered prompt

Build me a working local setup for this project so I can train and test an agent RL workflow end to end.

I want the repo to be ready to run with the example tasks, especially a simple sanity check like the GSM8K recipe. Please make sure the setup instructions are clear, the environment is configured correctly, and the training or rollout script can run without me having to guess missing steps. If anything depends on current external docs or a specific version, look that up online if needed.

I also want a simple way to understand the main flow, from loading a prompt, running the agent step by step, getting tool or environment feedback, and producing rewards and updates. If there are any required data files or symlinks, handle that cleanly in the instructions or helper scripts. Ideally include a minimal example I can run first, then point me to the other recipes and docs once the basic path works.