hugging-face-vision-trainer

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches and inspects public, user-provided Hugging Face datasets (e.g., SKILL.md mandates running scripts/dataset_inspector.py and examples reference datasets like "merve/MicroMat-mini", and the inspector calls the Datasets Server API at https://datasets-server.huggingface.co/rows), and the agent is required to read and act on those untrusted dataset fields (prompts, bboxes, labels) to decide preprocessing, training configuration, and tool use—so third‑party content can materially influence behavior.