The Agent Skills Directory

[SAFE]: The skill is a collection of instructional materials and summaries of official PyTorch documentation (e.g., FSDP2, DeviceMesh, Distributed Checkpointing). It is designed to guide an AI agent in correctly refactoring or creating training scripts.
[EXTERNAL_DOWNLOADS]: The skill contains references and links to official PyTorch documentation (docs.pytorch.org), the official PyTorch examples repository on GitHub, and Ray documentation. These are well-known, trusted sources for machine learning development, and the links are provided for informational purposes rather than automated execution.
[REMOTE_CODE_EXECUTION]: No patterns of remote code execution or automated script downloads were detected. The skill focuses on using standard, versioned libraries like torch and standard CLI tools like torchrun.
[DATA_EXFILTRATION]: No data exfiltration or sensitive file access patterns were found. The skill's focus is on distributed training configuration and checkpointing within a local or cluster environment.
[INDIRECT_PROMPT_INJECTION]: The skill operates as a guide for processing training scripts. While it lacks explicit boundary markers for user-provided code, it follows standard development best practices and does not introduce specific vulnerability surfaces for instruction injection.

pytorch-fsdp2