Dispatching work to an SSH/SLURM host the user has already configured: find out what the host actually offers, stage the files, submit and let the approval modal do its work, poll .result() from a later cell until the job is terminal, harvest the outputs, then write down what you learned about the host so the next session starts from it. Every submit puts a modal in front of the user and, once approved, spends their allocation — a string of failed submits costs their attention and their compute — so the recipe is shaped around landing the first one. It registers no SSH provider and grants access to no host.
Whether any of it works depends on the user's configuration, their credentials, the scheduler and allocation state, what software the remote actually has, and the approvals. Submitting a job spends real resources, so the recipe insists on validating the result and on explicit intent, and refuses to treat a queued command as a success.
A Skill is a directory of files, so installing one is copying that directory
somewhere an agent looks. With Node 18+ and git on PATH (npm resolves a
github: spec through it), and nothing cloned:
npx github:PKU-YuanGroup/OpenAI4S install remote-compute-ssh --target claude
--target claude writes to ~/.claude/skills, claude-project to
./.claude/skills, openai4s to <data_dir>/user-skills, and --dir <path>
to anywhere you name. The resolved absolute path is printed before anything is
written there, and --dry-run stops at that plan. A reinstall refuses to
overwrite a copy you have edited or one it did not install, and uninstall
removes only the files it wrote. For a recipe included in the npm release,
npx @pku-yuangroup/openai4s-skills install remote-compute-ssh --target claude
installs the published copy. The npm catalog can differ from this repository.
Without Node, take the directory itself — and turn it into a .zip if an upload
field wants one. The tarball is the whole repository (over 100 MB), and the pipe
is a POSIX shell recipe (macOS, Linux, WSL) that extracts only this directory:
curl -L https://codeload.github.com/PKU-YuanGroup/OpenAI4S/tar.gz/refs/heads/main \
| tar -xz --strip-components=2 OpenAI4S-main/skills/remote-compute-ssh
python3 -m zipfile -c remote-compute-ssh.zip remote-compute-ssh
The click-through form of that same download is the
repository zip,
which is also the route to take on Windows: extract it and copy
skills/remote-compute-ssh/ out
(unzip main.zip 'OpenAI4S-main/skills/remote-compute-ssh/*' unpacks only that
directory). If you already run OpenAI4S there is nothing to install — the wheel
ships every bundled Skill, and a bundled Skill takes precedence over a
same-named copy in <data_dir>/user-skills. Targets, provenance, and what the
installer refuses to do:
tools/skills-installer/.
| File | Responsibility |
|---|---|
SKILL.md |
The runbook for host.compute and the control kernel: why these calls belong in the repl tool rather than the python tool, how to read the compute details doc and how much discovery is left, finding a working environment activation, staging inputs, submitting directly or through SLURM, polling .result() until the job is terminal, harvesting, cancellation and recovery, and updating the host notes afterwards. |