The Agent Skills Directory

[SAFE]: The skill provides idiomatic PyTorch code snippets for training models. All operations, including gradient management, device placement, and mixed precision, follow industry-standard best practices.
[DATA_EXFILTRATION]: No network operations or sensitive data access were detected. The skill uses torch.save to write model checkpoints to the local directory, which is standard behavior for training scripts.
[REMOTE_CODE_EXECUTION]: No remote script downloads or dynamic code execution patterns are present. All logic is contained within static Python snippets.
[PROMPT_INJECTION]: The content is purely technical and does not contain instructions that attempt to manipulate the AI agent's behavior or override system constraints.

pytorch-training-loop