mmoore114/marl-trading-system — reverse-engineered prompt

Build me a simple end to end stock trading research project for US equities. I want it to pull historical price data from an API key in a local env file, create useful trading features, train a single reinforcement learning trading agent, then run a backtest on a held out period and give me clear results.

Please make it easy to control from one config file so I can change the stock universe, date range, factors, reward setup, and training settings without digging through code. The trading environment should use sensible constraints so it does not overtrade or take unrealistic positions.

At the end, I want a smooth pipeline I can run step by step, data download, feature creation, training, and backtesting, with logs plus a nice HTML performance report and an equity curve chart. Keep it reproducible and organized for experimentation, and if there are any old or placeholder multi agent pieces, keep the working version focused on the single agent baseline. Look up current docs online if you need to.