tao-train-depth-anything-v2
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides detailed documentation and YAML configurations for computer vision tasks using the NVIDIA TAO toolkit.
- [EXTERNAL_DOWNLOADS]: The skill references and fetches pretrained model weights from Meta's public file storage (fbaipublicfiles.com) and official NVIDIA container images from nvcr.io. These are well-known and legitimate sources for the skill's purpose.
- [COMMAND_EXECUTION]: Instructions include standard Docker commands for model training and inference. Documentation explicitly recommends running containers with non-root user IDs to maintain host-side security, which is a defensive best practice.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied annotation files and images. While these are entry points for external data, the structured nature of the data (annotations) and the use of specialized ML tooling within containers significantly limit the risk of instruction-based exploitation.
Audit Metadata