cuopt-multi-objective-exploration
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODEEXTERNAL_DOWNLOADS
Full Analysis
- [NO_CODE]: The skill consists entirely of Markdown instructions, evaluation data, and metadata. It does not include any scripts (Python, JavaScript, shell) or binaries, which eliminates the possibility of direct command execution or local file system abuse.
- [EXTERNAL_DOWNLOADS]: The
skill-card.mdandSKILL.mdfiles reference official NVIDIA documentation (docs.nvidia.com) and the official NVIDIA cuOpt examples repository on GitHub. These are recognized as trusted vendor resources. - [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to interpret user-supplied problem descriptions (e.g., procurement tradeoffs) to identify objectives and constraints for optimization solves.
- Ingestion points: User-provided natural language descriptions of optimization problems, such as those illustrated in
evals/evals.json. - Boundary markers: The workflow instructions do not define specific delimiters or instructions to ignore embedded commands within the input data.
- Capability inventory: The skill orchestrates mathematical solves performed by external tools defined in
cuopt-numerical-optimization-apiandcuopt-routing-api-python. - Sanitization: The skill provides logic for mathematical formulation but does not prescribe string sanitization for input data.
Audit Metadata