hugging-face-jobs

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow can ingest outsider-authored free text via user-provided Hub dataset contents (e.g., scripts like generate-responses.py / cot-self-instruct.py load an input dataset from the Hub and then pass its text fields into the job’s LLM generation context), which is not authored by the operating user.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly shows passing remote script URLs to hf_jobs() and even cites examples such as https://raw.githubusercontent.com/huggingface/trl/main/trl/scripts/sft.py (and Hugging Face raw dataset URLs) which are fetched at job runtime and executed as the job script, so remote content can directly execute code and control the agent's behavior.