KameniAlexNea/ludo-king-ai ? reverse-engineered prompt

Build me a Python project for a Ludo AI that can play the game using reinforcement learning.

I want a full Ludo rules engine with legal move checking, captures, blockades, safe squares, extra turns, and win detection, then wrap it in a Gymnasium style environment so I can train an agent on it. The app should expose masked actions, shaped rewards, and a board observation that the model can learn from. Please include a training script that uses Stable Baselines3 with a custom feature extractor, saves checkpoints, logs to TensorBoard, and can run self play against simple built in opponent strategies.

Also add a few helper tools so I can evaluate the agent, run small tournaments, and compare different strategies. Make it easy to tweak the reward settings and network settings in one place. If you need to, look up the current docs for Gymnasium and Stable Baselines3 so the training code matches the latest APIs.