tao-train-ocdnet

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

• Third-party content exposure detected (medium risk: 0.30). In tao-train-ocdnet, the runtime workflow ingests outsider-authored free text from dataset tarballs when used as datalist/list inputs (e.g., train_datasets/train.tar.gz via dataset.train_dataset.data_path), because such archives are extracted and then used to construct dataloaders from image/label mappings.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's deploy docs include docker run commands that pull and run the external container image nvcr.io/nvidia/tao/tao-toolkit:6.26.3-deploy, which is fetched at runtime and executes remote code inside the container.