huggingface
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWEXTERNAL_DOWNLOADS
Full Analysis
- External Content Loading (LOW): The examples use
AutoModel.from_pretrainedandpipeline, which download model weights and configuration files from the Hugging Face Hub. While Hugging Face is a trusted source, loading third-party models can pose a risk if the model files (e.g., pickle-based tensors) contain malicious code. The risk is mitigated by the library's increasing use of thesafetensorsformat. - Indirect Prompt Injection Surface (LOW): The skill is designed to process untrusted text data for tasks like summarization and question-answering. This exposes the agent to indirect prompt injection where malicious instructions could be embedded in the processed text. However, the skill does not provide the agent with high-privilege capabilities (like file system writes or arbitrary command execution) that could be triggered by such an injection.
- Network Operations (INFO): The 'Inference API' section demonstrates using
requeststo interact withapi-inference.huggingface.co. This is a legitimate and expected use of the library's services.
Audit Metadata