cuopt-numerical-optimization-api
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The Python reference model
assets/python/mps_solver/model.pyincludes a utility to download theair05.mps.gzbenchmark problem from the Zuse Institute Berlin's MIPLIB repository (https://miplib.zib.de/). This is a well-known academic repository for mathematical programming instances, and the download is used as a legitimate benchmark for the solver's performance. - [DATA_INGESTION]: The skill features multiple ingestion points for external optimization data, including
assets/python/mps_solver/model.py(viaProblem.readMPS),assets/c/mps_solver/mps_solver.c(viacuOptReadProblem), and thecuopt_clitool. These tools parse standard mathematical programming formats (MPS/LP). While these ingestion points represent an attack surface for Indirect Prompt Injection, the risk is minimal as the parsers follow strict format specifications and the primary purpose of the skill is numerical optimization. - Ingestion points:
assets/python/mps_solver/model.py,assets/c/mps_solver/mps_solver.c, and the CLI API documentation. - Boundary markers: None (parsing relies on strict file format specifications).
- Capability inventory: Subprocess calls and command execution are limited to building C examples and running the cuOpt solver via Python or CLI.
- Sanitization: Relies on the internal validation and parsing logic of the NVIDIA cuOpt library.
Audit Metadata