tao-run-on-kubernetes
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses standard Kubernetes commands (
kubectl) and shell scripts to perform environment checks and job management. These are documented as preflight steps and are within the expected scope of a Kubernetes execution platform. - [CREDENTIALS_UNSAFE]: The skill references several credentials (e.g.,
NGC_KEY,ACCESS_KEY,SECRET_KEY,HF_TOKEN) but correctly instructs the user to provide them via environment variables or secret management practices rather than hardcoding them. The benchmark report correctly identifies that these are passed as environment variables to pods, which is a standard Kubernetes operational pattern. - [EXTERNAL_DOWNLOADS]: The skill documents the use of external packages (
nvidia-tao-sdk[kubernetes]) and Helm charts for the NVIDIA GPU Operator. These are from well-known and trusted sources (NVIDIA's official registries and PyPI). - [DATA_EXFILTRATION]: No patterns of data exfiltration were detected. The skill supports S3 storage for job inputs and outputs, which is a legitimate and documented feature for ML workflows.
- [PROMPT_INJECTION]: No prompt injection or behavior override patterns were found in the instructions.
Audit Metadata