What does this skill do, and when should you use it?
This skill guides agents through repeated single-objective cuOpt solves to trace and interpret a Pareto frontier. It first optimizes each objective independently to build a payoff table, then sweeps either normalized weighted sums or ε-constraints. Candidate solutions are collected, deduplicated, filtered for dominance, and interpreted through exchange rates and knee points. The workflow covers LP, MILP, and QP models, and can also be applied to routing tradeoffs.
Defines and checks genuinely conflicting objectives; solves each objective independently and records a payoff table; sweeps normalized weighted sums or ε-constraint right-hand sides; runs cuOpt for each parameter combination and records objective values and solutions; removes dominated and duplicate points and sorts the survivors into a frontier; reads duals as local exchange rates for LP/QP linear ε-constraints and highlights knee regions.
- An operations researcher needs options balancing cost and service level but has no agreed weighting.
- An optimization team wants a complete coverage-versus-cost frontier and chooses ε-constraint sweeps.
- A MILP practitioner suspects weighted sums are hiding non-convex frontier regions.
- A planning or investment model needs to compare return, risk, budget, or other objectives using payoff-table ranges.
- A routing user wants to explore distance, vehicle-count, and time tradeoffs with the same frontier workflow.
How do you install this skill?
- Before execution, verify that cuOpt, the referenced API/formulation skills, and compatible versions are available; check solver status, gaps, time limits, and infeasibility.
- The skill does not choose weights or a final plan for the user; frontier completeness, knee identification, and exchange-rate interpretation still require review, especially for MILP and non-Optimal results.
- No Chinese examples or explicit mainland-China reachability guidance are provided; verify access in advance if the environment depends on overseas cuOpt services or documentation.
NVIDIA cuOpt
Install the specific skill through the repository README's skills CLI flow:
npx skills add nvidia/skills --skill cuopt-multi-objective-exploration --yes
The README does not specify a manual destination for this individual skill; the CLI prompts for the installation target.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/cuopt-multi-objective-exploration" ~/.claude/skills/
rm -rf "$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:
- Explore the Pareto frontier between cost and service level in cuOpt, build a payoff table first, then use ε-constraint sweeps and explain the knee.
With the skill loaded, ask an agent for a multi-objective cuOpt analysis, for example: “Explore the Pareto frontier between cost and service level in cuOpt, build a payoff table first, then use ε-constraint sweeps and explain the knee.” The skill supplies workflow guidance; concrete solver calls depend on the cuOpt API skills.
What are this skill's strengths and limitations?
- Separates a single weighted optimum from the broader tradeoff frontier.
- Uses a payoff table to establish sweep bounds and normalization scales.
- ε-constraints can recover non-convex regions missed by weighted sums.
- Requires reporting tradeoffs, knee points, solve status, and the selected weights or ε values.
- ε-constraint exploration requires more solves and bookkeeping.
- Weighted sums reach only the convex hull and may produce sparse MILP points.
- MILP and quadratic-constraint cases do not provide usable duals; MILP cannot use a PDLP warm start.
- The supplied source describes an instructional workflow and does not document bundled scripts or a test suite.
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 Multi-Objective Explorer this page ✓ NVIDIA · Official | 56 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| cuOpt Numerical Optimization Formulation ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| Portfolio AI Readiness Scan ✓ Anthropic · Official | 46 · Use with care | ★ 40k | 1mo ago | Apache-2.0 |
| Probability Analysis — Event Probability Analyst | 60 · Recommended | ★ 76 | 28d ago | MIT |
| Ad Experiment Designer | 59 · Recommended | ★ 3k | 13d ago | Apache-2.0 |
The skill explicitly contrasts weighted sums with ε-constraints: weighted sums suit a quick convex sketch, while ε-constraints are preferred for MILP, potentially non-convex frontiers, or completeness.
How did FollowSkills review this skill?
The skill confines itself to orchestrating existing cuOpt solves, avoids destructive actions, and says not to make the tradeoff decision for the user; the skill card also says not to place secrets in prompts, logs, or outputs. Deductions apply because user confirmation, data-flow details, permission boundaries, rollback, and dependency-security controls are not specified.
The document is internally coherent on payoff tables, weighted sums, epsilon constraints, dominance filtering, MILP time limits, dual interpretation, and non-optimal status reporting. Because this is a static review without committed key-path tests or CI coverage for the skill, the static ceiling applies and the score is limited to 9.
Activation and non-activation conditions, applicable LP/MILP/QP cases, and the choice between weighted sums and epsilon constraints are reasonably clear. Deductions apply for incomplete input/output contracts, solver-version and environment requirements, limited non-fit boundaries, no Chinese usage guidance, and reliance on cuOpt plus other API/formulation skills.
The skill has structured frontmatter, a version, Apache-2.0 licensing, an author team, tags, interface references, a benchmark, a skill card, and NVIDIA maintenance provenance. Deductions apply because the recommended Instructions and Examples sections are missing, the author format fails the static check, and there is no clear changelog, named maintenance responsibility, skill-specific update path, or local dependency-installation guidance.
It provides a directly reusable workflow covering single-objective anchoring, frontier sweeps, filtering, exchange rates, knee identification, and deferring the final decision to the user; the committed benchmark reports measured results across five tasks. Deductions apply because the benchmark is only a static report with a small one-attempt sample, and Codex effectiveness is 63%, so static evidence cannot establish broadly directly usable outputs.
SKILL.md, evals.json, and BENCHMARK.md provide auditable workflow descriptions, evaluation tasks, expected behaviors, and recorded results, supporting limited review. Deductions apply because there is no skill-specific test suite, CI reproduction path, raw execution artifact, or independent cross-source corroboration; static review cannot justify a higher score.
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 →