tao-train-single-step
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local helper scripts (e.g.,
scripts/list_tao_platforms.py,scripts/resolve_tao_image.py) to dynamically discover available compute platforms and resolve appropriate Docker images. These scripts are part of the vendor's automation toolkit and operate within the local environment. - [CREDENTIALS_UNSAFE]: The skill facilitates the intake of credentials required for training on specific platforms (e.g., S3, Slurm, or Kubernetes). The process is scoped to requested platforms and models, aligning with the expected administrative function of a training deployment tool.
- [SAFE]: Comprehensive analysis detected no signs of prompt injection, obfuscation, unauthorized remote code execution, or data exfiltration. The skill's behavior matches its documentation and the trusted context of the NVIDIA-TAO vendor resources.
Audit Metadata