skills/nvidia/skills/tao-run-platform/Gen Agent Trust Hub

tao-run-platform

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill installs the nvidia-tao-sdk[all] Python package from an environment-specific index. This is a vendor-owned package required for the skill's primary function.
  • [COMMAND_EXECUTION]: The skill performs shell commands to interact with job submission systems including Brev CLI, SLURM (sbatch), Docker, and Kubernetes (kubectl). It also executes local helper scripts for platform discovery and image resolution.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied parameters to build job specifications and commands. This is an inherent attack surface for job execution skills, which the skill addresses through structured spec construction and a dedicated entrypoint builder.
  • [DYNAMIC_EXECUTION]: The skill generates a container entrypoint that includes an inlined, Base64-encoded script runner runtime to handle data input/output within the execution environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 09:09 PM