hugging-face-jobs

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow (submit Hugging Face Jobs via hf_jobs() with script inline code) is designed to load and process user-selected Hugging Face Hub datasets/models, which are outsider-authored free text at runtime (e.g., prompts/messages from load_dataset(src_dataset_hub_id, split="train") and seed datasets in load_dataset(args.seed_dataset, split="train")).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill includes explicit examples that pass remote script URLs to the hf_jobs "script" parameter (which will be fetched and executed at runtime), for example: https://raw.githubusercontent.com/huggingface/trl/main/trl/scripts/sft.py, https://huggingface.co/datasets/uv-scripts/synthetic-data/raw/main/cot-self-instruct.py, https://huggingface.co/datasets/uv-scripts/dataset-stats/raw/main/finepdfs-stats.py, and https://huggingface.co/datasets/uv-scripts/vllm/raw/main/generate-responses.py, which means these external URLs can control executed code during job runtime.