---

name: openmc description: "Guide OpenMC Python and native Monte Carlo particle-transport workflows: install and build the runtime, construct models, prepare nuclear data and depletion, inspect tallies/results, and diagnose optional solver integrations." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

OpenMC operating skill

Use this skill when a task mentions OpenMC, continuous-energy or multigroup Monte Carlo transport, CSG reactor/fusion geometry, cross_sections.xml, statepoint/tally files, depletion chains, MGXS, random ray, CMFD, or the openmc.lib C API. This is a router, not a replacement for the detailed workflow references.

First classify the task

• Install, compile, run XML inputs, configure data paths, OpenMP/MPI, or diagnose an executable/library failure: read setup-runtime.
• Build materials, surfaces/regions/cells/universes/lattices, sources, settings, plots, or XML-only models: read model-geometry.
• Create filters/tallies or inspect statepoint, summary, track, restart, and plot output: read tallies-results.
• Validate cross-section/chain files, process ENDF/ACE/HDF5, use MGXS/MicroXS, or run depletion/decay/R2S: read nuclear-data-depletion.
• Use random ray, CMFD, weight windows, C API/library mode, C++/CTest, DAGMC, libMesh, NCrystal, or other optional native integrations: read advanced-solvers.

For a request spanning routes, establish setup/data gates first, construct the model next, then run transport and inspect outputs. Do not route an output-file problem to model construction merely because the model created the file.

Honest readiness gates

Report the gates separately:

1. import openmc and the base Python dependencies work.
2. An openmc executable is available for transport and subprocess workflows.
3. import openmc.lib loads a compiled shared library for C API/library mode.
4. cross_sections.xml parses and every referenced data file exists and is compatible with the requested energy mode.
5. Optional MPI/DAGMC/libMesh/NCrystal/strict-FP features are enabled and verified in the native build.

An XML export proves only Python model serialization. It does not prove a transport run, native library, cross-section coverage, or tally correctness. Use troubleshooting when a gate is ambiguous.

Minimal package check

For a supported Python environment, install the public package and its focused test tools as needed, then run:

python -m pip install openmc
python -c "import openmc; print(openmc.__version__)"

A source build is a separate CMake workflow; read setup-runtime before assuming that a Python install supplies the executable or libopenmc. Avoid downloading nuclear data or changing shell startup files without explicit user approval. Keep data and generated inputs in a dedicated working directory.

Shared operating rules

• Capture the OpenMC version, run mode, energy mode, particle type, data index, executable/library paths, and output directory before changing a workflow.
• Prefer small API/XML checks and bounded unit tests before transport, depletion, or full regression suites. Follow the repository's test guidance for OMP_NUM_THREADS and strict reproducibility when comparing results.
• Use explicit Path values and validate XML/HDF5 inputs before interpreting a physics or geometry error. Keep missing data, missing native artifacts, and invalid model inputs as distinct diagnoses.
• Never claim MPI, GPU, DAGMC, libMesh, NCrystal, or C API support from a Python import alone. Optional features are verified only by their native build and a safe feature-specific check.
• If the source checkout or package version has changed, read repo-provenance.md and use a refresh workflow rather than silently relying on stale operating knowledge.

Common references

• Read troubleshooting.md for cross-cutting install/import, data/config, CLI/API, native-build, and result-file failures.
• Read testing-and-verification.md for bounded native-test selection, data-dependent gates, and result expectations.
• Read repo-provenance.md before a refresh or when reconciling a skill claim with a different OpenMC revision.