Baba88611/detroit-ai-player ? reverse-engineered prompt

Build me a Python tool that lets an AI play through Detroit Become Human by reading each chapter as a decision tree and choosing between options on its own.

I want it to support OpenAI compatible models like GPT, Claude, DeepSeek, or a local endpoint, and let me switch models with a simple config. It should also let me choose different personas, like a default one or a more cautious or machine like personality, so I can see how the same story changes with different behavior. Keep the story data separate from the runner, and make sure the AI only sees the player facing scene text and choices, not the hidden state or consequences.

Please include a simple command line runner that can step through chapters, keep conversation context, update game state, and save the results so I can compare runs later. If you need current API details, look up the latest docs online.