The Agent Skills Directory

[SAFE]: The skill serves as a static documentation resource for PyTorch FSDP development. All included content is technical in nature and derived from established official documentation.
[SAFE]: Dependencies are limited to standard, reputable machine learning libraries including torch and transformers.
[INDIRECT_PROMPT_INJECTION]: The skill processes large documentation files (e.g., references/common-patterns.md) to provide context for the agent. This represents a standard information retrieval surface.
Ingestion points: The agent reads technical documentation from the references/ directory to answer queries.
Boundary markers: Content is naturally delimited by Markdown formatting and structured patterns.
Capability inventory: The skill enables the agent to assist in code development, sharding strategies, and debugging based on the provided reference material.
Sanitization: No specific input sanitization is implemented, which is standard for static reference skills.
[SAFE]: The documentation includes explicit security warnings regarding the use of the pickle module for distributed communication (e.g., recv_object_list), correctly identifying and advising against the risks of arbitrary code execution when unpickling untrusted data.

pytorch-fsdp