tao-finetune-clip

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

• Third-party content exposure detected (medium risk: 0.30). The only runtime LLM-ingested outsider-authored free text path is user-supplied prompt/caption content provided via spec inputs (e.g., inference.text_file: prompts.txt and caption/prompt files), which is loaded into the CLIP text encoder without first selecting a specific trusted item.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's deploy examples and runtime actions reference and pull the external container image nvcr.io/nvidia/tao/tao-toolkit:6.26.3-deploy (used in docker run commands), which is fetched at runtime and executes remote code required for the gen_trt_engine/evaluate/inference flows.