sol-skilllisted

Conventions and tooling for ASU's Sol supercomputer, built around the `solx` CLI. Use whenever a task is happening on Sol - the user mentions Sol or ASU Research Computing, or is clearly on their Sol account (a Sol /scratch path, an sbatch/interactive job, a login/compute node). It covers renewing /scratch files Sol has flagged for deletion (purge/inactivity warnings) via `solx keep` and where to store datasets and model caches; requesting and managing SLURM jobs (the `solx job` interactive-allocation lifecycle, sbatch for batch, GPU and partition/QOS choice, why a job is pending, fairshare-aware and time-aware job management); installing software without sudo (module load, uv for Python, tinytex for LaTeX); reaching a Sol compute-node service like Jupyter from a laptop browser; detecting login-vs-compute nodes and choosing where to run heavy I/O (the DTN, a compute node, or a batch job); and transferring data to and from Sol. Not for generic SLURM/HPC on other clusters (Phoenix, NERSC, ...), cloud GPUs, or p
Shu-Wan/solx · ★ 3 · AI & Automation · score 68

Install: claude install-skill Shu-Wan/solx

# 🌵 Sol skills Official doc: <https://docs.rc.asu.edu/>. That official doc is authoritative; these notes are just a cache. ## What this skill helps with This skill teaches an AI coding assistant how to operate on ASU's Sol supercomputer the way a careful human user would. The everyday job and scratch operations are driven through **`solx`**, a small CLI this skill installs and prefers; the rest is situational guidance for Sol's native tools. Concretely, it covers: - **Driving `solx`** - the CLI for daily Sol work: request and manage interactive allocations (`solx job start/list/jump/time/stop`) and renew flagged `/scratch` files (`solx keep`). The skill installs it on first use and prefers it. - **Detecting the environment** - recognizing whether the agent is running on a Sol login/compute node or on a laptop, and branching behavior accordingly. - **Storage decisions** - putting datasets, model weights, and caches under `/scratch/$USER` and keeping `/home` for config; never asking for `sudo`. - **Scratch retention** - renewing flagged files and directory trees with `solx keep`, using unified or legacy warning CSVs and the `[keep]` config block; interpreting skipped paths and owner-grouped permission failures; refusing to bulk-`touch` `/scratch`. - **Getting software onto Sol without sudo** - `module load` for what the cluster already provides, `uv` for Python interpreters and envs, R's `tinytex` for LaTeX, `~/.local`/`~/opt` for everything else.