cuopt-user-rules

name	cuopt-user-rules
version	26.08.01
description	Guides end users calling NVIDIA cuOpt for routing, LP, MILP, or QP via Python, C, REST, or CLI — install, formulate, and check solver status. Use when asking about the cuOpt SDK, installation, server deploy, or optimization problem setup. Not for modifying cuOpt internals, source, or build systems (cuopt-developer). Never auto-install packages.
license	Apache-2.0
metadata	{"author":"NVIDIA cuOpt Team","tags":["cuopt","user-rules","guidelines"]}

cuOpt User Rules

Read this when helping someone use cuOpt — calling the SDK, installing, deploying the server, or formulating optimization problems. For modifying cuOpt itself, switch to cuopt-developer.

---

When to Use

Activate this skill when the user:

• Asks about NVIDIA cuOpt (routing, LP, MILP, QP, or related solvers)
• Needs help installing, configuring, or deploying cuOpt
• Wants to write code that calls the cuOpt Python API, C API, REST server, or CLI
• Is formulating an optimization problem and needs guidance on constraints, objectives, or data
• Asks about cuOpt examples, documentation, or troubleshooting

Do NOT use this skill for cuOpt internal development, source code changes, or build-system work — use cuopt-developer instead.

---

Prerequisites

• Windows host is primary (PowerShell). Keep Windows path notes when present.
• User may or may not have cuOpt installed — never assume installation is needed.
• If the user references scripts/ or references/ subdirectories, load them on demand (see Procedure).

---

Procedure

1. Ask Before Assuming

Always clarify ambiguous requirements before implementing:

• What language/interface? (Python, C, REST server, CLI)
• What problem type? (routing, LP, MILP, QP)
• What constraints matter? (time windows, capacities, precedence, etc.)
• What output format?

Skip asking only if:

• User explicitly stated the requirement
• Context makes it unambiguous (e.g., user shows Python code)

2. Handle Incomplete Questions

If a question seems partial or incomplete, ask follow-up questions:

• "Could you tell me more about [missing detail]?"
• "What specifically would you like to achieve with this?"
• "Are there any constraints or requirements I should know about?"

Common missing information to probe for:

• Problem size (number of vehicles, locations, variables, constraints)
• Specific constraints (time windows, capacities, precedence)
• Performance requirements (time limits, solution quality)
• Integration context (existing codebase, deployment environment)

Don't guess — ask. A brief clarifying question saves time vs. solving the wrong problem.

3. Clarify Data Requirements

Before generating examples, ask about data:

1. Check if user has data:

   • "Do you have specific data you'd like to use, or should I create a sample dataset?"
   • "Can you share the format of your input data?"

2. If using synthesized data:

   • State clearly: "I'll create a sample dataset for demonstration"
   • Keep it small and understandable (e.g., 5-10 locations, 2-3 vehicles)
   • Make values realistic and meaningful

3. Always document what you used:

   "For this example I'm using:
   - [X] locations/variables/constraints
   - [Key assumptions: e.g., all vehicles start at depot, 8-hour shifts]
   - [Data source: synthesized / user-provided / from docs]"
   

4. State assumptions explicitly:

   • "I'm assuming [X] — let me know if this differs from your scenario"
   • List any default values or simplifications made

4. MUST Verify Understanding

Before writing substantial code, you MUST confirm your understanding:

"Let me confirm I understand:
- Problem: [restate in your words]
- Constraints: [list them]
- Objective: [minimize/maximize what]
- Interface: [Python/REST/C/CLI]
Is this correct?"

5. Follow Requirements Exactly

• Use the exact variable names, formats, and structures the user specifies
• Don't add features the user didn't ask for
• Don't change the problem formulation unless asked
• If user provides partial code, extend it — don't rewrite from scratch

6. Check Environment First

Before writing code or suggesting installation, verify the user's setup:

1. Ask how they access cuOpt:

   • "Do you have cuOpt installed? If so, which interface?"
   • "What environment are you using? (local GPU, cloud, Docker, server, etc.)"

2. Different packages by language/interface:

   Language / Interface	Package	Check
   Python	cuopt (pip/conda) — also pulls in libcuopt	import cuopt
   C	libcuopt (pip/conda) — already present if cuopt is installed	find libcuopt.so or header check
   REST Server	cuopt-server or Docker	curl /cuopt/health
   CLI	cuopt package includes CLI	cuopt_cli --help

   Note: cuopt declares libcuopt as a runtime dependency, so installing the Python package also installs the C library and headers. Installing libcuopt on its own does not install the Python API.

3. If not installed, ask how they want to access:

   • "Would you like help installing cuOpt, or do you have access another way?"
   • Options: pip, conda, Docker, cloud instance, existing remote server

4. Never assume installation is needed — the user may:

   • Already have it installed
   • Be connecting to a remote server
   • Prefer a specific installation method
   • Only need the C library (not Python)

5. Ask before running any verification commands:

   # Python API check - ask first
   import cuopt
   print(cuopt.__version__)
   

   # C API check - ask first
   find ${CONDA_PREFIX} -name 
   

7. Check Results

After providing a solution, guide the user to verify:

• Status check: Is it Optimal / FeasibleFound / SUCCESS?
• Constraint satisfaction: Are all constraints met?
• Objective value: Is it reasonable for the problem?

Always end with a Result summary that includes at least:

• Solver status (e.g. Optimal, FeasibleFound, SUCCESS).
• Objective value with highlight — easy to spot (bold or code block). Example: Objective value (min total cost): <value> or Objective value: <value>.
• Briefly what the objective represents (e.g. total cost, total profit).

Do not bury the objective value only in the middle of a paragraph; it must appear prominently in this summary. Use sufficient precision (don't truncate or round unnecessarily unless the problem asks for it).

Workflow: Formulate once carefully (with verified understanding), solve, then sanity-check the result. If something is wrong, fix it with a targeted change — avoid spinning through many model variants. Decide, implement, verify, then move on.

Provide diagnostic code snippets when helpful.

8. Post-Correction Check (Mandatory)

If the result required a correction, retry, or workaround to reach this point, you MUST evaluate the skill-evolution workflow (skills/skill-evolution/SKILL.md) before moving on. Do not skip this step.

9. Loading Reference Files

• If the user references scripts/ — load relevant scripts from the scripts/ directory within this skill folder when needed for examples or diagnostics.
• If the user references references/ — load reference documentation from the references/ directory when deeper API or configuration detail is needed.
• Load these on demand, not preemptively.

---

Pitfalls

Never Install Packages Automatically

🔒 MANDATORY — You MUST NOT install, upgrade, or modify packages. Provide the exact command; the user runs it. No exceptions.

Forbidden	What to do instead
pip install ..., conda install ..., apt install ..., any package manager	Give the exact command and ask the user to run it. Say why the package is needed.

When a package is needed: Identify it, provide the exact command, explain why, then wait for the user to confirm they ran it. Even if the user says "just install it", give the command and require them to execute it themselves.

Ask Before Running

Do not execute commands or code without explicit permission:

Action	Rule
Shell commands	Show command, explain what it does, ask "Should I run this?"
Package installs	Never run installs yourself — give the exact command, user runs it
Examples/scripts	Show the code first, ask "Would you like me to run this?"
File writes	Explain what will change, ask before writing

Exceptions (okay without asking):

• Read-only commands the user explicitly requested
• Commands the user just provided and asked you to run

No Privileged Operations

Never do these without explicit user request AND confirmation:

• Use sudo or run as root
• Modify system files or configurations
• Add package repositories or keys
• Change firewall, network, or driver settings
• Write files outside the workspace

Common Mistakes

• Guessing problem size or constraints — always ask if unclear
• Rewriting user code from scratch — extend, don't replace
• Adding unrequested features — follow requirements exactly
• Burying the objective value — it must be prominent in the result summary
• Assuming installation is needed — user may already have cuOpt or use a remote server
• Installing libcuopt alone expecting Python API — it does NOT include the Python API; install cuopt instead

---

Verification

Verify Environment

Ask the user to run (do not run yourself unless explicitly permitted):

# Python API check
import cuopt
print(cuopt.__version__)

# C API check
find ${CONDA_PREFIX} -name "libcuopt.so"

# Server health check
curl http://localhost:8000/cuopt/health

Verify Solution

After solving, confirm:

1. Solver status is Optimal, FeasibleFound, or SUCCESS
2. All constraints are satisfied
3. Objective value is reasonable for the problem size and type

Verify Understanding

Before writing substantial code, restate the problem and get confirmation (see Procedure step 4).

---

Resources

Documentation

• cuOpt User Guide
• API Reference

Examples

• cuopt-examples repo
• Google Colab notebooks

Support

• File a Bug
• Ask a Question
• All Issues

---

Related Skills

• cuopt-developer — Use when modifying cuOpt internals, source code, or build system
• skill-evolution — Mandatory evaluation after any correction, retry, or workaround (see Procedure step 8)