tao-train-dino

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

• Third-party content exposure detected (medium risk: 0.30). In references/tao-deploy-dino.skill_info.yaml, the deploy workflow ingests externally supplied COCO/classmap text files by downloading and extracting dataset.test_data_sources.json_file and dataset.infer_data_sources.classmap at runtime, but those inputs are user-provided artifacts rather than first-party trusted prose.

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

• Potentially malicious external URL detected (high risk: 0.80). The deploy examples invoke a remote Docker image (nvcr.io/nvidia/tao/tao-toolkit:6.26.3-deploy) via docker run, which would fetch and execute remote code at runtime (references/tao-deploy-dino.md), so this URL is a runtime external dependency that executes code.