hugging-face-jobs

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow centers on submitting a user-provided Python script string to the Hugging Face Jobs MCP tool (hf_jobs("uv"/"run", {...})), so the agent will ingest arbitrary outsider-authored free text (the script) without needing to select a specific item first.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly permits passing a remote "script" URL to hf_jobs (e.g., "https://raw.githubusercontent.com/huggingface/trl/main/trl/scripts/sft.py") and specifies pulling/runtime images (e.g., "ghcr.io/astral-sh/uv:python3.12-bookworm-slim", "vllm/vllm-openai:latest", "hf.co/spaces/lhoestq/duckdb"), which will be fetched and executed at job runtime and therefore constitute runtime external code execution.