llnl

Lessons learned from dftracer annotation sessions — real errors, root causes, and exact fixes. Loaded by each per-file annotation subagent at startup. Updated by the pipeline recipe after every session (Step 8).

Interactive dftracer annotation pipeline. Clones the app, builds it, annotates all source files using the clang-based MCP annotation tools (clang_annotate_project / clang_annotate_file + clang_syntax_check + clang_lint_annotations), builds the annotated…

Sweep ANY ice4hpc application across a configuration space (input x node-scale x processes-per-node) collecting full-feature dftracer traces — PAPI, GPU (ROCProfiler or CUPTI), variorum power, node utilization, MPI — into a validated, compacted,…

System profile for matrix (LLNL NVIDIA H100 + Intel Sapphire Rapids, TOSS 4, Slurm) — the toolchain that actually works, the three partitions and their very different limits and backlogs, where CUPTI really lives, the PAPI ceiling, and the two permission…

System profile for Tuolumne (LLNL AMD MI300A / Cray PE) — module load order, Flux scheduler, ROMIO/MPICH-IO hints, Lustre striping, ROCm/PyTorch environment setup, and known build/run pitfalls

The dftracer PyPI package itself (C core + CMake build) — package layout, dependency graph (dftracer-utils, pybind11, pydftracer), install sources (PyPI release vs GitHub develop via pip), and session-local vs shared-venv install rules. Load this skill before…

IOR-specific knowledge: build quirks, annotation pitfalls, ROMIO tuning on VAST NVMe storage, and optimal dftracer run configuration for IOR 4.0.0. Load this skill whenever working with IOR.

ScaFFold (LBANN PyTorch benchmark, 3D U-Net on procedurally-generated fractals). Python/PyTorch DDP app running on Tuolumne MI300A. Environment setup, shared venv pattern for AI/ML workloads, MPI4PY Cray linkage, ROCm paths.

Update or add SUNDIALS expected-output answer files after intentional test output changes. Use to refresh .out files embedded in examples, update the sundials-codes/answers repository or test/answers submodule, consume GitHub Actions or Jenkins output_files…

Review SUNDIALS pull requests using the repository’s correctness, numerical, API, documentation, testing, portability, and maintainability review patterns. Use for GitHub PR reviews or when asked to review changed SUNDIALS source, bindings, build files,…

Build, install, and test SUNDIALS from source as an end user or as a SUNDIALS developer. Use when a request involves configuring CMake, selecting compilers/options (MPI, GPU backends, Fortran), building/installing, running CTest, enabling dev/unit tests, or…

Fetch and apply GitHub pull-request review comments locally, especially when the task includes reconciling new review traffic, following reply chains from specific reviewers/authors, and keeping a scratch notes file that maps comment IDs to fixes.

Generate nanobind C++ wrapper code for user-supplied callback functions in the sundials4py Python bindings. Use this skill whenever adding a new user-supplied function (callback) to any sundials4py submodule (arkode, cvodes, idas, kinsol, core/sundials),…

Guide SUNDIALS users to the right package, time-stepping module, nonlinear/linear solver strategy, and initial tuning choices. Use when a request asks which SUNDIALS package to use (CVODE, CVODES, IDA, IDAS, ARKODE, KINSOL), whether to choose explicit vs…

Add a new SUNDIALS module (e.g., NVECTOR_*, SUNMATRIX_*, SUNLINSOL_*, SUNNONLINSOL_*, SUNMEMORY_*, or a new shared component) including source/header layout, CMake wiring, enable/disable options, exported CMake targets, installed component registration, and…

Use when preparing commits, reviewing git history for conventions, or writing pull request summaries for this repository.

Use when working on LCTIO development setup, editable installs, linting, formatting, type checking, or packaging.

Use when updating Sphinx documentation, changing docs/source/conf.py, adding Sphinx extensions, or validating documentation builds for this repository.

Use when modifying or debugging lctio.retiff, TIFF rewrite behavior, or the retiff CLI.

Use when writing or updating pytest coverage for TIFF handling, CLI behavior, or other tests in this repository.

Guide for building the tool when asked by the user or testing a code change that requires a new build of the tool

Guide for understanding the structure and content of HTML reports generated by the tool

Guide for writing tests in the project

Guide for understanding the tool's functionality and usage in error tracking

Build and install the AMS library (github.com/LLNL/AMS) from source on an HPC cluster via CMake. Use this whenever the user wants to compile, configure, build, or install AMS, or is choosing between build variants (with/without RabbitMQ, GPU, MPI, Caliper,…

Maintain a CHANGELOG.md in Keep a Changelog format: record new features, changes, and fixes under an Unreleased section, and cut releases. Use whenever a user-facing change ships or the user says "update the changelog", "add a changelog entry", "record this…

Build and maintain a living architecture wiki for a codebase: one markdown article per module/subsystem plus an index that maps how they fit together. Use whenever the user wants to document, map, or reason about the code's architecture — "map the codebase",…

Commands for running and monitoring jobs on a Flux (flux-framework) HPC scheduler: submitting jobs, checking whether a job is running, reading job output and exit codes, and cancelling jobs. Use this whenever a task involves submitting, monitoring,…

Implementation-oriented local `dmx-learn` fitting skill for coding, fitting, diagnosing, and using models on local or in-memory data once the problem structure is known. Prefer explicit `dmx.stats` estimators for ordinary non-Bayesian work and use…

Use when adding or updating tests, or when validating dmx-learn changes against the repo's CI quality gates. Covers pytest layout, the shared stats test harness, torch test constraints, and the exact Black, isort, mypy, pylint, and pydocstyle checks used in…

Main entry point for local `dmx-learn` modeling requests. Use to scope the problem, infer the observation structure, keep the workflow on local data, and route detailed implementation to narrower repo-local skills or references. Default to explicit…

Use when modifying Python source, utilities, examples, or distribution code in dmx-learn. Applies the repo's enforced coding standards from CI: Poetry workflows, Black line length 88, isort black profile, typed function signatures for mypy, pylint-friendly…

Instructions for building Smith

C++ coding style and naming conventions for Smith (primarily src/smith), including formatting via the CMake `style` target.

Instructions for creating unit tests for Smith

Instructions for building Axom

C++ coding style and naming conventions for Axom (primarily src/axom), including formatting via the CMake `style` target.

Instructions for creating unit tests for Axom

Place, move, or review `BOBA_CALI_MARK`, `BOBA_CALI_BEGIN`, `BOBA_CALI_SWITCH`, and `BOBA_CALI_END` in BoBa code. Use when refactoring backend wrappers, splitting algorithms across helper layers, or fixing profiling ownership so CALI scopes match the code…

Compile and run BoBa with Caliper enabled, capture `CALI_CONFIG=runtime-report` output, and interpret which rows correspond to BoBa profiling regions. Use when profiling a test, tutorial, or miniapp, or when deciding between `BOBA_ENABLE_CALIPER`,…

Debug BoBa code with `checkpoint()`, `always_checkpoint()`, `boba_print(...)`, and debug builds such as `make <target> BOBA_DEBUG=1`. Use when narrowing where a failure occurs, adding temporary runtime breadcrumbs, or printing variable state while a test or…

Add a new benchmark experiment to Benchpark (experiments/{benchmark}/experiment.py) or extend an existing one with programming models (+openmp/+cuda/+rocm), scaling studies (strong/weak/throughput), workloads, or Caliper support. Use this whenever someone…

Use Kosh to persist results and files, query them later, and build custom loaders, transformers, and operators for data extraction and slice-aware processing.

Gather unresolved GitLab merge request comments for the current branch with `grape -d review --printUnresolvedComments`, fix trivial feedback, verify the resulting code changes, create a commit, and summarize how each comment was addressed. Use when working…