tao-train-grounding-dino
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill provides configuration schemas and documentation for using the NVIDIA TAO Toolkit.
- [EXTERNAL_DOWNLOADS]: The skill references official NVIDIA container images (nvcr.io) and BERT weights from HuggingFace. These are well-known and trusted sources for model weights and containerized tools.
- [COMMAND_EXECUTION]: The instructions and scripts focus on executing the 'grounding_dino' CLI tool within a TAO container. The commands are standard for machine learning training, evaluation, and inference workflows.
- [INDIRECT_PROMPT_INJECTION]: While the model processes text prompts for object detection, these inputs are used as detection targets by the BERT text encoder and do not influence the agent's control flow or security boundaries.
Audit Metadata