The Agent Skills Directory

[Environment and Tool Usage]: The skill interacts with the Hugging Face Jobs infrastructure and utilizes standard machine learning libraries (Transformers, Timm, Datasets). These operations are consistent with its primary purpose of vision model training.
[Credential Management]: The skill follows secure practices by instructing users to pass sensitive tokens via job secrets (e.g., using $HF_TOKEN placeholders or get_token()) rather than hardcoding credentials. It explicitly documents that tokens are required for pushing results to the Hub, ensuring user control over authentication.
[External References]: The skill downloads and references configuration and scripts from trusted sources like Hugging Face and well-known GitHub repositories. These interactions are documented clearly and follow established patterns for cloud-based machine learning workflows.
[Data Handling]: The provided scripts for dataset inspection and training utilize established APIs (such as the Datasets Server API) to validate data formats efficiently before allocating GPU resources, which serves as a safeguard against common training failures.
[Persistence Mechanisms]: The skill uses the Hugging Face Hub for model persistence. This is a standard and intended use case for the platform, ensuring that training results from ephemeral cloud environments are saved to the user's specific namespace.

huggingface-vision-trainer