tao-finetune-cosmos-reason

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

• Third-party content exposure detected (high risk: 0.85). At runtime, the skill ingests outsider-authored free text from the user-selected dataset annotation JSON (e.g., dataset.annotation_path / custom.*.annotation_path), which contains fields like question/gt/response that are loaded and used during evaluation; these are not authored by the operating user.