william-dawson/hpc-fugaku
Facility-specific HPC workflow skills for Fugaku. Install the hpc base plugin first for MCP tools.
Use when building, compiling, or linking software on Fugaku — choosing a compiler toolchain (Fujitsu vs LLVM vs GCC), MPI, BLAS/LAPACK, cross-compilation for A64FX, and verifying a build actually produces working parallel code.
Use when the user wants to set up, configure, or troubleshoot Fugaku — SSH access, the embedding endpoint for docs search (RAG), or the ~/.hpc-agent/fugaku.json file. Also use when fugaku tools fail with connection or embedding errors.
Interactive demo of Fugaku — walks through configuration, facility info, live cluster status, docs search, filesystem access, and a real job end to end.
Use when the user asks about the status, progress, output, history, or failure of jobs on Fugaku, or about queue and node availability. Also use this any time you (the agent) need to check on or wait for a job you submitted, even without a fresh user request.
Use to answer questions about Fugaku itself — hardware, partitions, storage, software, policies, or where to get help. Ground answers in the bundled guide and live tool output rather than prior knowledge.
Use when the user wants to run one or more shell commands on Fugaku's login node, or when another workflow needs run_command_on_cluster for a live check not covered by a safer typed tool. Requires an exact command preview and explicit user permission before execution.
Use when the user asks to make a result on Fugaku reproducible, shareable, or turned into a notebook/script — or proactively suggest it after a chunk of exploratory job/file work that produced something worth preserving. Builds a Jupyter notebook that replays the real workflow via hpc_agent_core.client, not a hand-copied SSH script.
Use when the user wants to run, submit, or launch a job on Fugaku. Covers partition/queue selection, JobSpec construction, submission, and known gotchas specific to this facility.