skills/nvidia/skills/cuopt-install/Gen Agent Trust Hub

cuopt-install

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install software from trusted vendor sources, including the NVIDIA PyPI index (pypi.nvidia.com), official NVIDIA Conda channels, and the official NVIDIA Docker Hub repository. These resources are verified as belonging to the vendor (nvidia).
  • [COMMAND_EXECUTION]: Instructions include standard shell commands for package management (pip, conda, docker) and system diagnostics (nvidia-smi, nvcc, curl). These commands are necessary for the installation and verification of the cuOpt library.
  • [DYNAMIC_EXECUTION]: The verification documentation includes a step to compile a static C source code template using gcc. This is used solely to verify the installation of C headers and libraries, presenting minimal risk as it does not incorporate untrusted runtime data into the compilation process.
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions involve checking local environment variables and system outputs (such as CUDA version and GPU architecture) to determine the appropriate installation package. This intended ingestion of environmental metadata is a standard part of the software selection process.
  • [METADATA_POISONING]: The skill includes benchmark and signature files that confirm the integrity and performance of the skill. No deceptive metadata or instruction overrides were found in the headers or descriptive fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 11:28 PM
Security Audit — agent-trust-hub — cuopt-install