About
This skill enhances Reinforcement Learning (RL) trading agents by providing them with critical account-level context, including cumulative profit and loss, rolling win rates, and current drawdown percentages. By incorporating these features into the observation space and applying a quadratic drawdown penalty in the reward function, the model learns to adapt its behavior based on past performance—becoming more conservative during losing streaks and leveraging gains more effectively. It is specifically designed for high-performance vectorized environments where capital preservation is as crucial as signal generation, offering a robust framework for training risk-aware financial AI.
