Stable Baselines3 RL Assistant

What it does

Built around Stable Baselines3 (SB3) 2.8, this skill lets Claude drive an end-to-end reinforcement learning workflow:

• Algorithm selection across PPO, A2C, SAC, TD3, DDPG, DQN and HER
• Custom gymnasium.Env templates plus check_env() validation rules (image dtype, channel order, Discrete start constraints)
• Vectorized environments (DummyVecEnv vs SubprocVecEnv) and off-policy gradient_steps=-1 tuning
• Callbacks: EvalCallback, CheckpointCallback, reward thresholds, custom callback scaffolding
• Model persistence, VecNormalize statistics, evaluation and video recording
• Learning-rate schedules, Dict observations with MultiInputPolicy, TensorBoard logging

Bundled scripts/ templates and references/ guides back up each section.

Who it's for

• Researchers and grad students reproducing or prototyping RL experiments
• Engineers applying single-agent RL to robotics, game AI or scheduling
• Developers new to the SB3 API who want a safe, standard workflow

Examples

1. "Wrap my inventory simulator as a Gymnasium env and validate it with check_env" → generates the env class and validation script
2. "Continuous control task — pick between SAC and TD3 and write the training loop" → algorithm comparison plus training code with callbacks and TensorBoard
3. "Evaluate my saved PPO agent over 10 episodes and record a video" → evaluate_policy and VecVideoRecorder snippets

Install the skill I found on Claude Skill Mart.
Copy the skills/stable-baselines3 folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/stable-baselines3/.
When it's done, tell me in one line what this skill can do.