The Agent Skills Directory

[SAFE] (SAFE): The script is a legitimate reinforcement learning training template using the pufferlib library.
[COMMAND_EXECUTION] (SAFE): The script executes training logic locally using PyTorch. Parameters for environment names, batch sizes, and learning rates are handled securely via argparse.
[EXTERNAL_DOWNLOADS] (SAFE): The script utilizes standard machine learning libraries (torch, pufferlib) and integrations for logging services like Weights & Biases and Neptune. These are trustworthy and typical for ML workflows.
[CREDENTIALS_UNSAFE] (LOW): The script allows providing a Neptune API token via command line arguments (--neptune-token). While no secrets are hardcoded, passing tokens via CLI arguments can expose them to process monitors or shell history. Users should be advised to use environment variables for sensitive tokens.

pufferlib