The Agent Skills Directory

Secret Management and Authentication: The skill enforces best practices for handling Hugging Face authentication tokens. It instructs the agent to use encrypted secrets via the HF_TOKEN environment variable rather than hardcoding credentials or using less secure environment configurations. This follows industry standards for credential management in remote execution environments.
Remote Code Execution Context: The skill is designed to execute Python scripts on Hugging Face's managed cloud infrastructure. This functionality is the primary purpose of the skill and is implemented through standardized interfaces like the hf_jobs MCP tool and huggingface_hub library. The execution of remote scripts is restricted to well-known and trusted sources, such as Hugging Face's official GitHub and Hub repositories.
Data Processing and Safety Boundaries: The included scripts ingest external data from the Hugging Face Hub for tasks like batch inference and dataset analysis. To mitigate the surface for indirect prompt injection, the scripts utilize the apply_chat_template method from the Transformers library, which is a recognized best practice for separating untrusted user data from model instructions.
Dependency Management: The skill uses PEP 723 inline dependency headers to manage its software environment. The identified dependencies consist of reputable open-source libraries (e.g., PyTorch, Transformers, Polars, vLLM) commonly used for large-scale data and machine learning tasks.

huggingface-jobs