tao-train-oneformer

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md specifies the training/inference actions consume dataset files (e.g., annotations.json, label_map.json, images.tar.gz) via spec_overrides/data_sources without any prior selection step inside the workflow, so outsider-authored content in those uploaded dataset artifacts can be read at runtime.