huggingface-best

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The skill fetches public Hugging Face benchmark datasets and leaderboards at runtime from https://huggingface.co/api/datasets?... and https://huggingface.co/api/datasets/<namespace>/<repo>/leaderboard, where the returned JSON/metadata (e.g., dataset descriptions/tags and leaderboard entries) is outsider-authored content that can be ingested into the agent’s LLM context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill performs runtime curl requests to Hugging Face APIs (e.g., https://huggingface.co/api/datasets?filter=benchmark:official&limit=500, https://huggingface.co/api/datasets///leaderboard, and https://huggingface.co/api/models/org/model1) to fetch leaderboards and model metadata that directly drive the agent's model-selection logic and output, so these external endpoints are required and control the agent's behavior.