What does this skill do, and when should you use it?
This skill guides AI agents in using NVIDIA cuOpt for linear programming (LP), mixed-integer linear programming (MILP), and beta quadratic programming (QP). It selects guidance for Python, C/C++, or command-line workflows and helps classify the problem and variable types. It also covers using cuOpt as a solver backend for AMPL, GAMS/GAMSPy, PuLP, JuMP, Pyomo, and CVXPY. Version 26.08.00 is published by the NVIDIA cuOpt Team under Apache-2.0.
Selects a Python, C/C++, or CLI reference based on the user's interface; classifies LP, MILP, and QP from the objective and variable types; maps wording such as counts, on/off choices, proportions, and rates to integer, binary, or continuous variables; explains QP restrictions including minimization only, continuous variables, and positive-semidefinite Q; documents availability limits for dual values, reduced costs, and PDLP warmstarts; diagnoses infeasibility, unboundedness, slow solves, rejected QP maximization, and poor QP scaling; and introduces time_limit, mip_relative_gap, mip_absolute_tolerance, and log_to_console settings.
- A Python developer building and solving an LP or MILP with cuOpt.||||A systems engineer embedding cuOpt in a C or C++ application and needing interface-specific guidance.
- A user with MPS files who wants to solve from the command line.
- A team with an existing AMPL, GAMS/GAMSPy, PuLP, JuMP, Pyomo, or CVXPY model that wants cuOpt as the backend instead of porting the model.
- A user modeling portfolio variance, least squares, or regularized regression who needs to check whether the problem fits cuOpt's continuous QP rules.
How do you install this skill?
- This is a static review only; installation, compilation, solving, and the benchmark were not executed. The referenced Python/C/CLI materials and key-path tests should be supplied and reviewed before publication.
- QP is labeled beta while also described as production-capable for typical convex QP; compatibility, stability, and supported-use limits should be clarified.
- Users must confirm cuOpt, GPU availability, library paths, and documentation reachability in the target environment; no Chinese-language guidance or mainland-China reachability guarantee is provided.
- Results may be affected by infeasibility, unboundedness, numerical scaling, time limits, and MIP gaps; the skill guidance does not replace independent model and solution validation.
- Local filesystem
NVIDIA cuOpt
Install the individual skill from the NVIDIA/skills collection with the skills CLI:
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --yes
You can target Codex explicitly:
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent codex
The supplied source does not document cuOpt installation, version requirements, or hardware requirements.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/cuopt-numerical-optimization-api" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Use the cuOpt Python API to build and solve a production-planning MILP with integer variables.
After installation, state the interface and optimization task in the request, for example: “Use the cuOpt Python API to build and solve a production-planning MILP with integer variables.” If the interface is unclear, the skill directs the agent to ask first. If an AMPL, GAMS/GAMSPy, PuLP, JuMP, Pyomo, or CVXPY model already exists, identify that tool so cuOpt can be considered as its solver backend.
What are this skill's strengths and limitations?
- Covers LP, MILP, and beta QP with explicit problem-selection and variable-type rules.
- Provides guidance for Python, C/C++, and CLI interfaces.
- Includes troubleshooting guidance for infeasibility, unboundedness, performance, and QP numerical issues.
- Supports a documented backend path for several third-party modeling languages.
- QP is beta, supports minimization only and continuous variables, and requires a positive-semidefinite Q for a convex, well-posed problem.
- MILP has no dual values; duals are also unavailable with quadratic constraints.
- The supplied material documents guidance and references but does not include concrete API code, cuOpt installation instructions, or test results.
- The CLI workflow assumes MPS files, but the source does not explain how to create them.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| cuOpt Numerical Optimization API this page ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| cuOpt REST Server Deployment & Python Client ✓ NVIDIA · Official | 45 · Not recommended | ★ 3.6k | 13d ago | Apache-2.0 |
| cuOpt Python Routing ✓ NVIDIA · Official | 53 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| Codex CLI Builder ✓ OpenAI · Official | 55 · Use with care | ★ 28k | 4mo ago | — |
| NotebookLM Automation | 53 · Use with care | ★ 20k | 13d ago | MIT |
For users already working in AMPL, GAMS/GAMSPy, PuLP, JuMP, Pyomo, or CVXPY, the source explicitly recommends using cuOpt as the solver backend to minimize code changes. Otherwise, users can follow the cuOpt Python, C/C++, or CLI interface guidance.
How did FollowSkills review this skill?
The skill is scoped to cuOpt LP/MILP/QP modeling and solving, asks for clarification when the interface is unclear, and shows no credential access, exfiltration, destructive defaults, or excessive permissions. It does not document sensitive-data handling, dependency security, solver data flows, user confirmation, rollback, or result-risk controls, so 7 points are deducted.
The document gives interface routing, problem classification, QP constraints, common issues, and some C build examples, with limited diagnostic guidance for failures. However, the referenced Python/C/CLI references are not supplied, no key-path tests for this skill are shown, and the QP beta/production-capable wording is potentially unsafe. Under the static ceiling and with these gaps, it receives 8.
Audience and activation conditions are relatively clear, covering Python, C/C++, CLI, and LP/MILP/QP, with boundaries such as unsupported integer QP stated. Non-fit ranges, input/output contracts, compatibility, Chinese support, and mainland-China reachability are insufficiently documented; core documentation links depend on NVIDIA-hosted overseas pages, so 5 points are deducted.
The skill has a version, name, tags, Apache-2.0 license metadata, interface routing, troubleshooting guidance, and C assets. It lacks the recommended Instructions and Examples sections; the author format fails the stated schema finding; and there is no changelog, explicit maintenance/update path, complete installation dependency note, or visible Python/CLI examples. Seven points are deducted.
The stated goal is to guide users through LP/MILP/QP solving, and the classification rules, settings, and C assets provide potentially usable guidance. The included benchmark reports a pass across 9 tasks with correctness, discoverability, and effectiveness metrics, but it lacks reproducible task outputs, while the Python/C/CLI reference contents are unavailable here. Direct completeness and usability therefore remain uncertain, yielding 6 under static calibration.
Versioned source, C example assets, repository CI, and an attached benchmark summary provide limited audit evidence. The skill has no committed key-path test suite, and the benchmark is only a summary without task details or independent reproduction artifacts, so it does not meet the stronger corroboration threshold and receives 4.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →