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.py includes a utility to download the air05.mps.gz benchmark 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 (via Problem.readMPS), assets/c/mps_solver/mps_solver.c (via cuOptReadProblem), and the cuopt_cli tool. 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
Risk Level
SAFE
Analyzed
Aug 8, 2026, 02:33 AM
Security Audit — agent-trust-hub — cuopt-numerical-optimization-api