The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs standard, well-known machine learning libraries from the official Python Package Index (PyPI), including transformers, huggingface_hub, and datasets. These dependencies are necessary for the library's intended functionality and originate from trusted sources.
[REMOTE_CODE_EXECUTION]: The documentation covers the use of the trust_remote_code=True parameter for loading custom model architectures from the Hugging Face Hub. The skill correctly identifies this as a potential security risk and instructs the user to only enable it for code that has been manually reviewed.
[CREDENTIALS_UNSAFE]: No hardcoded credentials were found. The skill provides proactive security guidance for handling Hugging Face authentication tokens (HF_TOKEN), recommending the use of environment variables or secret managers and explicitly warning against committing tokens to version control or hardcoding them in scripts.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a data processing surface by handling external inputs (text, audio, vision) through model pipelines and tokenizers. This is an expected part of the library's function, and the skill includes no patterns that would exploit agent capabilities or bypass safety guardrails.

transformers