tao-train-single-step
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local scripts such as
list_tao_platforms.pyandresolve_tao_image.pyto gather configuration data about supported platforms and Docker images. These are standard operations for managing the TAO training environment. - [EXTERNAL_DOWNLOADS]: The instructions reference official NVIDIA NGC pretrained checkpoints and container images. Since the author of the skill is NVIDIA and these are official vendor resources, they are considered safe references for the intended training workflow.
- [CREDENTIALS_UNSAFE]: The workflow involves asking the user for credentials relevant to the selected compute platform (e.g., cloud backend). This is a necessary step for deployment, though it is handled through user interaction rather than hardcoded secrets.
Audit Metadata