Alee08/multiagent-rl-rm — reverse-engineered prompt

Build me a Python package for experimenting with multi agent reinforcement learning using Reward Machines.

I want to describe tasks in plain words, like go to A, then B, then C, and turn that into a validated reward machine file that training code can use. Include example worlds such as Frozen Lake and Office World, with simple maps, obstacles, goals, and two agents that can each have their own reward machine. Add runnable training scripts from the command line, using simple Q learning, plus a Python API example so researchers can import the environments and build their own experiments.

Please include a small command line tool that can generate reward machines from text, with a mock mode for offline tests and an OpenAI compatible local Ollama option. It should save JSON or YAML specs and catch invalid machines. Add basic tests, install instructions, Docker support, and a clear README with quickstart commands.