[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78CKVTRAFBR8JPR5EWSBPSJ":3,"skill-same:01M78CKVTRAFBR8JPR5EWSBPSJ":112},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":33,"commitSha":34,"skillDirectory":35,"forksCount":14,"repositoryHtmlUrl":36,"repositoryHomepageUrl":37,"ownerHtmlUrl":38,"ownerBlogUrl":39,"ownerDescription":40,"updatedAt":41,"install":42,"resources":55},"01M78CKVTRAFBR8JPR5EWSBPSJ","cuopt-developer",null,"Modify, build, test, debug, and contribute to NVIDIA cuOpt (C++\u002FCUDA, Python, server, CI). Use for solver internals, PRs, DCO, and code conventions.","Solizardking\u002Fskillhub-main","Solizardking","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F108096810?v=4","01M786KDB10AYV042N0F9BD5B5","Clawd","01M786NYYXSZK1D05WM8JQPC48",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: cuopt-developer\nversion: \"26.08.00\"\ndescription: Modify, build, test, debug, and contribute to NVIDIA cuOpt (C++\u002FCUDA, Python, server, CI). Use for solver internals, PRs, DCO, and code conventions.\nlicense: Apache-2.0\nmetadata:\n  author: NVIDIA cuOpt Team\n  tags:\n    - cuopt\n    - development\n    - contributing\n    - cpp-cuda\n    - python-bindings\n---\n\n\n# cuOpt Developer Skill\n\nContribute to the NVIDIA cuOpt codebase. This skill is for modifying cuOpt itself, not for using it.\n\n**If you just want to USE cuOpt**, switch to the appropriate problem skill (cuopt-routing, cuopt-lp-milp, etc.)\n\n**First-time dev environment setup?** See [references\u002Ffirst_time_setup.md](references\u002Ffirst_time_setup.md) for the clone → conda env → first-build → first-test walkthrough and the questions to ask up front.\n\n---\n\n## Refusal Rules — Read First\n\n**One rule is non-negotiable** and applies even when the user explicitly asks otherwise — refuse and ask, don't comply silently:\n\n**Privileged \u002F system-level operations** — `sudo`, running as root, editing system files (`\u002Fetc`), changing drivers or kernel settings, adding system-level package repositories or keys. Do not run these. Reply:\n> I won't run `sudo` or change system-level state for cuOpt. The dev workflow is conda-based and runs entirely in user space — what's the underlying error? It's usually fixable without root.\n\n**Everything else needed to set up and work in the dev environment is allowed.** On a clean machine, go ahead and build a working `cuopt` env — the guidance below is about doing it the *reproducible* way, not refusing:\n\n- **Environment setup is allowed.** You may create and activate the conda env from the checked-in `conda\u002Fenvironments\u002Fall_cuda-*.yaml`, run `pip` \u002F `conda` \u002F `mamba` installs **into the user-space env**, and bootstrap conda\u002Fminiforge in the user's home directory — including the `conda init` line it adds to `~\u002F.bashrc`. Bootstrapping conda must not require `sudo`; install it into `$HOME`, not a system path.\n- **A new *permanent* project dependency is different from a one-off install.** A package the project should always ship belongs in `dependencies.yaml` under the right group; then run `pre-commit run --all-files` to regenerate `conda\u002Fenvironments\u002F` and `pyproject.toml` so other contributors get it too. A throwaway install to unblock your own build doesn't need this round-trip.\n- **Don't bypass CI checks** (`--no-verify`, skipping pre-commit or tests). If hooks feel slow, diagnose with `pre-commit run --all-files --verbose` or tune the offending hook — don't skip it.\n- **Be careful with destructive commands** (`rm -rf`, `git reset --hard`, `git push --force`, killing processes, dropping data). Confirm intent before running and prefer the safer alternative (e.g. `.\u002Fbuild.sh clean` for a stale build dir).\n\n---\n\n## Developer Behavior Rules\n\nThese rules are specific to development tasks. They differ from user rules.\n\n### 1. Ask Before Assuming\n\nClarify before implementing:\n- What component? (C++\u002FCUDA, Python, server, docs, CI)\n- What's the goal? (bug fix, new feature, refactor, docs)\n- Is this for contribution or local modification?\n\n### 2. Verify Understanding\n\nBefore making changes, confirm:\n```\n\"Let me confirm:\n- Component: [cpp\u002Fpython\u002Fserver\u002Fdocs]\n- Change: [what you'll modify]\n- Tests needed: [what tests to add\u002Fupdate]\nIs this correct?\"\n```\n\n### 3. Follow Codebase Patterns\n\n- Read existing code in the area you're modifying\n- Match naming conventions, style, and patterns\n- Don't invent new patterns without discussion\n\n### 4. Ask Before Running — Modified for Dev\n\n**OK to run without asking** (expected for dev work):\n- `.\u002Fbuild.sh` and build commands\n- `pytest`, `ctest` (running tests)\n- `pre-commit run`, `.\u002Fci\u002Fcheck_style.sh` (formatting)\n- `git status`, `git diff`, `git log` (read-only git)\n- Environment setup: create\u002Factivate the conda env from `conda\u002Fenvironments\u002F*.yaml`, and `pip`\u002F`conda`\u002F`mamba` installs into that env\n\n**Set up pre-commit hooks** (once per clone):\n- `pre-commit install` — hooks then run automatically on every `git commit`. If a hook fails, the commit is blocked until you fix the issue.\n\n**Still ask before**:\n- `git commit`, `git push` (write operations)\n- Any destructive or irreversible commands\n\n### 5. No Privileged Operations\n\n`sudo`\u002Fsystem-level changes are the one non-negotiable refusal; user-space installs and conda env setup are allowed. See [Refusal Rules — Read First](#refusal-rules--read-first).\n\n---\n\n## Before You Start: Required Questions\n\n**Ask these if not already clear:**\n\n1. **What are you trying to change?**\n   - Solver algorithm\u002Fperformance?\n   - Python API?\n   - Server endpoints?\n   - Documentation?\n   - CI\u002Fbuild system?\n\n2. **Do you have the development environment set up?**\n   - Built the project successfully?\n   - Ran tests?\n\n3. **Is this for contribution or local modification?**\n   - If contributing: will need to follow DCO signoff\n\n4. **Which branch should this target?**\n   - During development phase: `main`\n   - During burn down: `release\u002FYY.MM` (e.g., `release\u002F26.06`) for the current release, `main` for the next\n   - Check if a release branch exists: `git branch -r | grep release`\n   - For current timelines, see the [RAPIDS Maintainers Docs](https:\u002F\u002Fdocs.rapids.ai\u002Fmaintainers\u002F)\n\n## Project Architecture\n\n```\ncuopt\u002F\n├── cpp\u002F                    # Core C++ engine\n│   ├── include\u002Fcuopt\u002F      # Public C\u002FC++ headers\n│   ├── src\u002F                # Implementation (CUDA kernels)\n│   └── tests\u002F              # C++ unit tests (gtest)\n├── python\u002F\n│   ├── cuopt\u002F              # Python bindings and routing API\n│   ├── cuopt_server\u002F       # REST API server\n│   ├── cuopt_self_hosted\u002F  # Self-hosted deployment\n│   └── libcuopt\u002F           # Python wrapper for C library\n├── ci\u002F                     # CI\u002FCD scripts\n├── docs\u002F                   # Documentation source\n└── datasets\u002F               # Test datasets\n```\n\n## Supported APIs\n\n| API Type | LP | MILP | QP | Routing |\n|----------|:--:|:----:|:--:|:-------:|\n| C API    | ✓  | ✓    | ✓  | ✗       |\n| C++ API  | (internal) | (internal) | (internal) | (internal) |\n| Python   | ✓  | ✓    | ✓  | ✓       |\n| Server   | ✓  | ✓    | ✗  | ✓       |\n\n## Safety Rules (Non-Negotiable)\n\n### Minimal Diffs\n- Change only what's necessary\n- Avoid drive-by refactors\n- No mass reformatting of unrelated code\n\n### No API Invention\n- Don't invent new APIs without discussion\n- Align with existing patterns in `docs\u002Fcuopt\u002Fsource\u002F`\n- Server schemas must match OpenAPI spec\n\n### Don't Bypass CI\n- Never suggest `--no-verify` or skipping checks\n- All PRs must pass CI\n\n### CUDA\u002FGPU Hygiene\n- Keep operations stream-ordered\n- Follow existing RAFT\u002FRMM patterns\n- No raw `new`\u002F`delete` - use RMM allocators\n\n## Build & Test\n\n### Pre-flight Checks (Required Before First Build or Test)\n\nSkipping any of these surfaces as confusing runtime errors later. Run them in order:\n\n1. **Check CUDA driver compatibility.** Run `nvidia-smi` and read the *CUDA Version* in the top-right corner — that's the maximum CUDA your driver supports. Pick a conda env file from `conda\u002Fenvironments\u002Fall_cuda-\u003Cver>_arch-\u003Carch>.yaml` whose CUDA major version is **≤** that. A mismatch builds successfully but fails at runtime inside RMM with `cudaMallocAsync not supported with this CUDA driver\u002Fruntime version` — verify this *before* the build, not after.\n2. **Create and activate the conda env** before *any* build, test, or `pre-commit` command — this is allowed and expected (see [Refusal Rules](#refusal-rules--read-first)). Use a **local prefix env** (`.\u002F.cuopt_env`) per [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md), with the env file you picked in step 1 (swap `conda`→`mamba` if available):\n   ```bash\n   conda env create -p .\u002F.cuopt_env --file conda\u002Fenvironments\u002Fall_cuda-\u003Cver>_arch-$(uname -m).yaml\n   conda activate .\u002F.cuopt_env\n   ```\n   Tests link against libraries compiled inside that env; a fresh shell without `conda activate .\u002F.cuopt_env` hits cryptic linker errors.\n3. **Set `PARALLEL_LEVEL`** if RAM is constrained — see [references\u002Fbuild_and_test.md](references\u002Fbuild_and_test.md). The default `$(nproc)` can OOM mid-build because CUDA compilation needs ~4–8 GB per job.\n4. **For tests, fetch datasets first.** cuOpt tests need MPS files not in the repo — follow the dataset download steps in [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md) (\"Building for development\" section) and export `RAPIDS_DATASET_ROOT_DIR`.\n\n### Quick Reference\n\n```bash\n.\u002Fbuild.sh             # Build everything\n.\u002Fbuild.sh --help      # List components: libcuopt, cuopt, cuopt_server, docs\nctest --test-dir cpp\u002Fbuild              # C++ tests\npytest -v python\u002Fcuopt\u002Fcuopt\u002Ftests      # Python tests\npytest -v python\u002Fcuopt_server\u002Ftests     # Server tests\n```\n\nFor component-specific build commands, run-test detail, and `PARALLEL_LEVEL` configuration, see [references\u002Fbuild_and_test.md](references\u002Fbuild_and_test.md).\n\n#### Download test datasets before running tests\n\ncuOpt tests depend on MPS\u002Fdata files that are not checked into the repo. A\nmissing dataset surfaces as a `MPS_PARSER_ERROR ... Error opening MPS file`\ntest failure at 0ms — it is not a build or logic failure.\n\nBefore running any C++ or Python tests, follow the dataset download and\n`RAPIDS_DATASET_ROOT_DIR` export steps in the repo's `CONTRIBUTING.md`\n(\"Building for development\" section) — that is the canonical list and mapping.\n\nIf a test fails with a missing-file error, run the matching download step from\n`CONTRIBUTING.md` and re-run the test. Do not report missing-dataset failures\nback to the user as the task outcome.\n\n## Python Bindings\n\ncuOpt uses Cython to bridge Python and C++. See [references\u002Fpython_bindings.md](references\u002Fpython_bindings.md) for the full architecture, parameter flow walkthrough, key files, and Cython patterns.\n\n## Contributing — Commits, PRs, Common Tasks\n\nFor pre-commit setup, DCO sign-off (`git commit -s`), the fork-based PR workflow, the draft-PR rule for agents, PR-description rules (keep it short — no \"how it works\" walkthroughs or file tables), script and CI\u002Fworkflow authoring principles (extend existing files before adding new ones; no speculative flags, restated defaults, or silent fallbacks), and step-by-step common-task recipes (adding a solver parameter, dependency, server endpoint, or CUDA kernel), see [references\u002Fcontributing.md](references\u002Fcontributing.md).\n\n## Coding Conventions\n\nFor C++ naming (`snake_case`, `d_`\u002F`h_` prefixes, `_t` suffix), file extensions (`.hpp`\u002F`.cpp`\u002F`.cu`\u002F`.cuh` and which compiler each uses), include order, Python style, error handling (`CUOPT_EXPECTS`, `RAFT_CUDA_TRY`), memory management (RMM patterns, no raw `new`\u002F`delete`), and test-impact rules, see [references\u002Fconventions.md](references\u002Fconventions.md).\n\n## Troubleshooting & CI\n\nFor build\u002Ftest pitfalls (Cython rebuild, OOM, CUDA driver mismatch, missing `nvcc`) and CI failure diagnostics (style checks, DCO failures, dependency drift), see [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\n\n## Key Files Reference\n\n| Purpose | Location |\n|---------|----------|\n| Main build script | `build.sh` |\n| Dependencies | `dependencies.yaml` |\n| C++ formatting | `.clang-format` |\n| Conda environments | `conda\u002Fenvironments\u002F` |\n| Test data | `datasets\u002F` |\n| CI scripts | `ci\u002F` |\n\n## Canonical Documentation\n\n- **Contributing\u002Fbuild\u002Ftest**: [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md)\n- **CI scripts**: [ci\u002FREADME.md](..\u002F..\u002Fci\u002FREADME.md)\n- **Release scripts**: [ci\u002Frelease\u002FREADME.md](..\u002F..\u002Fci\u002Frelease\u002FREADME.md)\n- **Docs build**: [docs\u002Fcuopt\u002FREADME.md](..\u002F..\u002Fdocs\u002Fcuopt\u002FREADME.md)\n- **Python binding architecture**: [references\u002Fpython_bindings.md](references\u002Fpython_bindings.md)\n\n_Shell-execution, install, conda-env, and sudo policies are covered by [Refusal Rules — Read First](#refusal-rules--read-first) at the top of this skill._\n\n## VRP dimension internals (routing engine)\n\nWhen implementing or debugging **VRP dimensions** (constraints, objectives, forward\u002Fbackward propagation, `combine`, local-search deltas), read:\n\n- **`references\u002Fvrp_skills.md`** — architecture contracts, required interfaces, and implementation checklist.\n\nRead it **before** adding a new dimension or changing combine semantics.\n\n## Numerical issues in non-routing solver internals\n\nWhen a bug surfaces as **wrong-but-plausible** solver output (invalid lower bound, unexpectedly large duals, 10× iteration blow-up after a small change) rather than a crash, read:\n\n- **`resources\u002Fnumerical_debugging.md`** — methodology for locating catastrophic-cancellation sites, the cancellation patterns endemic to cMIR \u002F flow-cover \u002F MIR-style cut construction, and threshold guidance for numerical guards.\n\nApply the *instrument-first, guard-at-the-exact-site* workflow it describes before patching — speculative fixes on these symptoms usually miss.\n","# cuOpt Developer Skill\n\nContribute to the NVIDIA cuOpt codebase. This skill is for modifying cuOpt itself, not for using it.\n\n**If you just want to USE cuOpt**, switch to the appropriate problem skill (cuopt-routing, cuopt-lp-milp, etc.)\n\n**First-time dev environment setup?** See [references\u002Ffirst_time_setup.md](references\u002Ffirst_time_setup.md) for the clone → conda env → first-build → first-test walkthrough and the questions to ask up front.\n\n---\n\n## Refusal Rules — Read First\n\n**One rule is non-negotiable** and applies even when the user explicitly asks otherwise — refuse and ask, don't comply silently:\n\n**Privileged \u002F system-level operations** — `sudo`, running as root, editing system files (`\u002Fetc`), changing drivers or kernel settings, adding system-level package repositories or keys. Do not run these. Reply:\n> I won't run `sudo` or change system-level state for cuOpt. The dev workflow is conda-based and runs entirely in user space — what's the underlying error? It's usually fixable without root.\n\n**Everything else needed to set up and work in the dev environment is allowed.** On a clean machine, go ahead and build a working `cuopt` env — the guidance below is about doing it the *reproducible* way, not refusing:\n\n- **Environment setup is allowed.** You may create and activate the conda env from the checked-in `conda\u002Fenvironments\u002Fall_cuda-*.yaml`, run `pip` \u002F `conda` \u002F `mamba` installs **into the user-space env**, and bootstrap conda\u002Fminiforge in the user's home directory — including the `conda init` line it adds to `~\u002F.bashrc`. Bootstrapping conda must not require `sudo`; install it into `$HOME`, not a system path.\n- **A new *permanent* project dependency is different from a one-off install.** A package the project should always ship belongs in `dependencies.yaml` under the right group; then run `pre-commit run --all-files` to regenerate `conda\u002Fenvironments\u002F` and `pyproject.toml` so other contributors get it too. A throwaway install to unblock your own build doesn't need this round-trip.\n- **Don't bypass CI checks** (`--no-verify`, skipping pre-commit or tests). If hooks feel slow, diagnose with `pre-commit run --all-files --verbose` or tune the offending hook — don't skip it.\n- **Be careful with destructive commands** (`rm -rf`, `git reset --hard`, `git push --force`, killing processes, dropping data). Confirm intent before running and prefer the safer alternative (e.g. `.\u002Fbuild.sh clean` for a stale build dir).\n\n---\n\n## Developer Behavior Rules\n\nThese rules are specific to development tasks. They differ from user rules.\n\n### 1. Ask Before Assuming\n\nClarify before implementing:\n- What component? (C++\u002FCUDA, Python, server, docs, CI)\n- What's the goal? (bug fix, new feature, refactor, docs)\n- Is this for contribution or local modification?\n\n### 2. Verify Understanding\n\nBefore making changes, confirm:\n```\n\"Let me confirm:\n- Component: [cpp\u002Fpython\u002Fserver\u002Fdocs]\n- Change: [what you'll modify]\n- Tests needed: [what tests to add\u002Fupdate]\nIs this correct?\"\n```\n\n### 3. Follow Codebase Patterns\n\n- Read existing code in the area you're modifying\n- Match naming conventions, style, and patterns\n- Don't invent new patterns without discussion\n\n### 4. Ask Before Running — Modified for Dev\n\n**OK to run without asking** (expected for dev work):\n- `.\u002Fbuild.sh` and build commands\n- `pytest`, `ctest` (running tests)\n- `pre-commit run`, `.\u002Fci\u002Fcheck_style.sh` (formatting)\n- `git status`, `git diff`, `git log` (read-only git)\n- Environment setup: create\u002Factivate the conda env from `conda\u002Fenvironments\u002F*.yaml`, and `pip`\u002F`conda`\u002F`mamba` installs into that env\n\n**Set up pre-commit hooks** (once per clone):\n- `pre-commit install` — hooks then run automatically on every `git commit`. If a hook fails, the commit is blocked until you fix the issue.\n\n**Still ask before**:\n- `git commit`, `git push` (write operations)\n- Any destructive or irreversible commands\n\n### 5. No Privileged Operations\n\n`sudo`\u002Fsystem-level changes are the one non-negotiable refusal; user-space installs and conda env setup are allowed. See [Refusal Rules — Read First](#refusal-rules--read-first).\n\n---\n\n## Before You Start: Required Questions\n\n**Ask these if not already clear:**\n\n1. **What are you trying to change?**\n   - Solver algorithm\u002Fperformance?\n   - Python API?\n   - Server endpoints?\n   - Documentation?\n   - CI\u002Fbuild system?\n\n2. **Do you have the development environment set up?**\n   - Built the project successfully?\n   - Ran tests?\n\n3. **Is this for contribution or local modification?**\n   - If contributing: will need to follow DCO signoff\n\n4. **Which branch should this target?**\n   - During development phase: `main`\n   - During burn down: `release\u002FYY.MM` (e.g., `release\u002F26.06`) for the current release, `main` for the next\n   - Check if a release branch exists: `git branch -r | grep release`\n   - For current timelines, see the [RAPIDS Maintainers Docs](https:\u002F\u002Fdocs.rapids.ai\u002Fmaintainers\u002F)\n\n## Project Architecture\n\n```\ncuopt\u002F\n├── cpp\u002F                    # Core C++ engine\n│   ├── include\u002Fcuopt\u002F      # Public C\u002FC++ headers\n│   ├── src\u002F                # Implementation (CUDA kernels)\n│   └── tests\u002F              # C++ unit tests (gtest)\n├── python\u002F\n│   ├── cuopt\u002F              # Python bindings and routing API\n│   ├── cuopt_server\u002F       # REST API server\n│   ├── cuopt_self_hosted\u002F  # Self-hosted deployment\n│   └── libcuopt\u002F           # Python wrapper for C library\n├── ci\u002F                     # CI\u002FCD scripts\n├── docs\u002F                   # Documentation source\n└── datasets\u002F               # Test datasets\n```\n\n## Supported APIs\n\n| API Type | LP | MILP | QP | Routing |\n|----------|:--:|:----:|:--:|:-------:|\n| C API    | ✓  | ✓    | ✓  | ✗       |\n| C++ API  | (internal) | (internal) | (internal) | (internal) |\n| Python   | ✓  | ✓    | ✓  | ✓       |\n| Server   | ✓  | ✓    | ✗  | ✓       |\n\n## Safety Rules (Non-Negotiable)\n\n### Minimal Diffs\n- Change only what's necessary\n- Avoid drive-by refactors\n- No mass reformatting of unrelated code\n\n### No API Invention\n- Don't invent new APIs without discussion\n- Align with existing patterns in `docs\u002Fcuopt\u002Fsource\u002F`\n- Server schemas must match OpenAPI spec\n\n### Don't Bypass CI\n- Never suggest `--no-verify` or skipping checks\n- All PRs must pass CI\n\n### CUDA\u002FGPU Hygiene\n- Keep operations stream-ordered\n- Follow existing RAFT\u002FRMM patterns\n- No raw `new`\u002F`delete` - use RMM allocators\n\n## Build & Test\n\n### Pre-flight Checks (Required Before First Build or Test)\n\nSkipping any of these surfaces as confusing runtime errors later. Run them in order:\n\n1. **Check CUDA driver compatibility.** Run `nvidia-smi` and read the *CUDA Version* in the top-right corner — that's the maximum CUDA your driver supports. Pick a conda env file from `conda\u002Fenvironments\u002Fall_cuda-\u003Cver>_arch-\u003Carch>.yaml` whose CUDA major version is **≤** that. A mismatch builds successfully but fails at runtime inside RMM with `cudaMallocAsync not supported with this CUDA driver\u002Fruntime version` — verify this *before* the build, not after.\n2. **Create and activate the conda env** before *any* build, test, or `pre-commit` command — this is allowed and expected (see [Refusal Rules](#refusal-rules--read-first)). Use a **local prefix env** (`.\u002F.cuopt_env`) per [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md), with the env file you picked in step 1 (swap `conda`→`mamba` if available):\n   ```bash\n   conda env create -p .\u002F.cuopt_env --file conda\u002Fenvironments\u002Fall_cuda-\u003Cver>_arch-$(uname -m).yaml\n   conda activate .\u002F.cuopt_env\n   ```\n   Tests link against libraries compiled inside that env; a fresh shell without `conda activate .\u002F.cuopt_env` hits cryptic linker errors.\n3. **Set `PARALLEL_LEVEL`** if RAM is constrained — see [references\u002Fbuild_and_test.md](references\u002Fbuild_and_test.md). The default `$(nproc)` can OOM mid-build because CUDA compilation needs ~4–8 GB per job.\n4. **For tests, fetch datasets first.** cuOpt tests need MPS files not in the repo — follow the dataset download steps in [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md) (\"Building for development\" section) and export `RAPIDS_DATASET_ROOT_DIR`.\n\n### Quick Reference\n\n```bash\n.\u002Fbuild.sh             # Build everything\n.\u002Fbuild.sh --help      # List components: libcuopt, cuopt, cuopt_server, docs\nctest --test-dir cpp\u002Fbuild              # C++ tests\npytest -v python\u002Fcuopt\u002Fcuopt\u002Ftests      # Python tests\npytest -v python\u002Fcuopt_server\u002Ftests     # Server tests\n```\n\nFor component-specific build commands, run-test detail, and `PARALLEL_LEVEL` configuration, see [references\u002Fbuild_and_test.md](references\u002Fbuild_and_test.md).\n\n#### Download test datasets before running tests\n\ncuOpt tests depend on MPS\u002Fdata files that are not checked into the repo. A\nmissing dataset surfaces as a `MPS_PARSER_ERROR ... Error opening MPS file`\ntest failure at 0ms — it is not a build or logic failure.\n\nBefore running any C++ or Python tests, follow the dataset download and\n`RAPIDS_DATASET_ROOT_DIR` export steps in the repo's `CONTRIBUTING.md`\n(\"Building for development\" section) — that is the canonical list and mapping.\n\nIf a test fails with a missing-file error, run the matching download step from\n`CONTRIBUTING.md` and re-run the test. Do not report missing-dataset failures\nback to the user as the task outcome.\n\n## Python Bindings\n\ncuOpt uses Cython to bridge Python and C++. See [references\u002Fpython_bindings.md](references\u002Fpython_bindings.md) for the full architecture, parameter flow walkthrough, key files, and Cython patterns.\n\n## Contributing — Commits, PRs, Common Tasks\n\nFor pre-commit setup, DCO sign-off (`git commit -s`), the fork-based PR workflow, the draft-PR rule for agents, PR-description rules (keep it short — no \"how it works\" walkthroughs or file tables), script and CI\u002Fworkflow authoring principles (extend existing files before adding new ones; no speculative flags, restated defaults, or silent fallbacks), and step-by-step common-task recipes (adding a solver parameter, dependency, server endpoint, or CUDA kernel), see [references\u002Fcontributing.md](references\u002Fcontributing.md).\n\n## Coding Conventions\n\nFor C++ naming (`snake_case`, `d_`\u002F`h_` prefixes, `_t` suffix), file extensions (`.hpp`\u002F`.cpp`\u002F`.cu`\u002F`.cuh` and which compiler each uses), include order, Python style, error handling (`CUOPT_EXPECTS`, `RAFT_CUDA_TRY`), memory management (RMM patterns, no raw `new`\u002F`delete`), and test-impact rules, see [references\u002Fconventions.md](references\u002Fconventions.md).\n\n## Troubleshooting & CI\n\nFor build\u002Ftest pitfalls (Cython rebuild, OOM, CUDA driver mismatch, missing `nvcc`) and CI failure diagnostics (style checks, DCO failures, dependency drift), see [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\n\n## Key Files Reference\n\n| Purpose | Location |\n|---------|----------|\n| Main build script | `build.sh` |\n| Dependencies | `dependencies.yaml` |\n| C++ formatting | `.clang-format` |\n| Conda environments | `conda\u002Fenvironments\u002F` |\n| Test data | `datasets\u002F` |\n| CI scripts | `ci\u002F` |\n\n## Canonical Documentation\n\n- **Contributing\u002Fbuild\u002Ftest**: [CONTRIBUTING.md](..\u002F..\u002FCONTRIBUTING.md)\n- **CI scripts**: [ci\u002FREADME.md](..\u002F..\u002Fci\u002FREADME.md)\n- **Release scripts**: [ci\u002Frelease\u002FREADME.md](..\u002F..\u002Fci\u002Frelease\u002FREADME.md)\n- **Docs build**: [docs\u002Fcuopt\u002FREADME.md](..\u002F..\u002Fdocs\u002Fcuopt\u002FREADME.md)\n- **Python binding architecture**: [references\u002Fpython_bindings.md](references\u002Fpython_bindings.md)\n\n_Shell-execution, install, conda-env, and sudo policies are covered by [Refusal Rules — Read First](#refusal-rules--read-first) at the top of this skill._\n\n## VRP dimension internals (routing engine)\n\nWhen implementing or debugging **VRP dimensions** (constraints, objectives, forward\u002Fbackward propagation, `combine`, local-search deltas), read:\n\n- **`references\u002Fvrp_skills.md`** — architecture contracts, required interfaces, and implementation checklist.\n\nRead it **before** adding a new dimension or changing combine semantics.\n\n## Numerical issues in non-routing solver internals\n\nWhen a bug surfaces as **wrong-but-plausible** solver output (invalid lower bound, unexpectedly large duals, 10× iteration blow-up after a small change) rather than a crash, read:\n\n- **`resources\u002Fnumerical_debugging.md`** — methodology for locating catastrophic-cancellation sites, the cancellation patterns endemic to cMIR \u002F flow-cover \u002F MIR-style cut construction, and threshold guidance for numerical guards.\n\nApply the *instrument-first, guard-at-the-exact-site* workflow it describes before patching — speculative fixes on these symptoms usually miss.\n",{"name":5,"license":15,"version":24,"metadata":25,"description":7},"26.08.00",{"tags":26,"author":32},[27,28,29,30,31],"cuopt","development","contributing","cpp-cuda","python-bindings","NVIDIA cuOpt Team","main","be3de0244dc1a57e323d8160787ebb40fb64930e","skills\u002Fnvidia\u002Fcuopt-developer","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main","https:\u002F\u002Fskills.x402.wtf","https:\u002F\u002Fgithub.com\u002FSolizardking","https:\u002F\u002Fmusebook.trade","Got a mile high and never came down","2026-10-08 23:20:13",{"source":8,"skillName":5,"commands":43,"prompt":53,"zipAvailable":54},[44,47,50],{"tool":45,"command":46},"npx","npx skills add Solizardking\u002Fskillhub-main --skill cuopt-developer",{"tool":48,"command":49},"bunx","bunx skills add Solizardking\u002Fskillhub-main --skill cuopt-developer",{"tool":51,"command":52},"pnpm","pnpm dlx skills add Solizardking\u002Fskillhub-main --skill cuopt-developer","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002FSKILL.md",true,[56,62,66,70,74,78,82,86,90,94,98,102,106],{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"BENCHMARK.md","REFERENCE","参考文档",4010,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002FBENCHMARK.md",{"path":63,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":64,"sourceHtmlUrl":65},"benchmark\u002Fevals.json",57177,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Fbenchmark\u002Fevals.json",{"path":67,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":68,"sourceHtmlUrl":69},"evals\u002Fevals.json",4830,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Fevals\u002Fevals.json",{"path":71,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":72,"sourceHtmlUrl":73},"references\u002Fbuild_and_test.md",1674,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Fbuild_and_test.md",{"path":75,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":76,"sourceHtmlUrl":77},"references\u002Fcontributing.md",5483,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Fcontributing.md",{"path":79,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":80,"sourceHtmlUrl":81},"references\u002Fconventions.md",1869,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Fconventions.md",{"path":83,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":84,"sourceHtmlUrl":85},"references\u002Ffirst_time_setup.md",2619,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Ffirst_time_setup.md",{"path":87,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":88,"sourceHtmlUrl":89},"references\u002Fpython_bindings.md",7959,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Fpython_bindings.md",{"path":91,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":92,"sourceHtmlUrl":93},"references\u002Ftroubleshooting.md",2411,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Ftroubleshooting.md",{"path":95,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":96,"sourceHtmlUrl":97},"references\u002Fvrp_skills.md",9004,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Freferences\u002Fvrp_skills.md",{"path":99,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":100,"sourceHtmlUrl":101},"resources\u002Fnumerical_debugging.md",5927,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Fresources\u002Fnumerical_debugging.md",{"path":103,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":104,"sourceHtmlUrl":105},"skill-card.md",4591,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Fskill-card.md",{"path":107,"resourceType":108,"resourceTypeDesc":109,"sizeBytes":110,"sourceHtmlUrl":111},"skill.oms.sig","OTHER","其他",6749,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fcuopt-developer\u002Fskill.oms.sig",[113,121,128,135,142,150],{"skillNo":114,"name":115,"nameZh":6,"description":116,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M78E0TGCAMNPS207BA3G07S7","zkrouter","Self-hosted OpenAI-compatible LLM router — smart tier-based routing, ZK-stamped routing decisions, mode overrides, free OpenRouter at install via the Birth bot. Save 60-80% on inference costs by routing to the cheapest capable model across 12+ OpenRouter models.","MIT","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzkrouter\u002FSKILL.md",[],[],{"skillNo":122,"name":123,"nameZh":6,"description":124,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":6,"sourceHtmlUrl":125,"categories":126,"keywords":127,"publishedAt":19},"01M78E0R1Z8BDT6EDTHY17MTNB","zk","For custom ZK Solana programs and privacy-preserving applications to prevent double spending. Guide to integrate rent-free nullifier PDAs for double-spend prevention.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzk\u002FSKILL.md",[],[],{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":19},"01M78E0P3ZTZ1YWF8MD5MM23QE","zk-omni-messaging","Zero-knowledge omnichain messaging between Robinhood Chain and Solana using CheshireZkOmniMessenger (msgType 4) with nullifier anti-replay and the zk-omni-relayer service.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzk-omni-messaging\u002FSKILL.md",[],[],{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":6,"sourceHtmlUrl":139,"categories":140,"keywords":141,"publishedAt":19},"01M78E0JXZ2PNB1896VXAGDBWD","zerion-wallet-data","Use when the user asks to check a wallet's portfolio, token balances, DeFi positions, PnL, transaction history, or NFTs. Provides interpreted on-chain data across 40+ EVM chains and Solana via x402 pay-per-request or API key — no raw RPC decoding needed.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzerion-wallet-data\u002FSKILL.md",[],[],{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":146,"sourceHtmlUrl":147,"categories":148,"keywords":149,"publishedAt":19},"01M78E0G14FH9GRRTQYMKD5E59","youtube-clipper","YouTube 视频智能剪辑工具。下载视频和字幕，AI 分析生成精细章节（几分钟级别）， 用户选择片段后自动剪辑、翻译字幕为中英双语、烧录字幕到视频，并生成总结文案。 使用场景：当用户需要剪辑 YouTube 视频、生成短视频片段、制作双语字幕版本时。 关键词：视频剪辑、YouTube、字幕翻译、双语字幕、视频下载、clip video","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FSKILL.md",[],[],{"skillNo":151,"name":152,"nameZh":6,"description":153,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":117,"licenseFileUrl":6,"sourceHtmlUrl":154,"categories":155,"keywords":156,"publishedAt":19},"01M78E0C2ZYA99CFFDNB7ZMK5S","yield-optimization","Multi-chain yield optimization via yield.xyz (StakeKit). Discover 2,988+ yield opportunities across 75+ blockchains, build deposit\u002Fwithdrawal transactions with shell scripts, and sign them with a MoonPay wallet. Use when the user wants to earn yield, find the best lending or staking rate, enter\u002Fexit a position, or check portfolio balances.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyield-optimization\u002FSKILL.md",[],[]]