Skip to content

Lawrencium Skill for Claude

Science IT publishes a Claude Code plugin that teaches Claude Code how to answer your Lawrencium and Einsteinium questions directly from the official Science IT documentation.

Once installed, whenever you ask a question about accounts, logging in, Slurm, software modules, Open OnDemand, or data transfer, Claude reads the current Markdown source of this documentation site and grounds its answer in it — instead of scraping the rendered website.

What this is

A skill packaged as a Claude Code plugin. It ships only a small set of instructions and a map of the documentation — no copy of the docs is bundled. The pages are fetched live from the lbnl-science-it/scienceit-docs GitHub repository as needed.

Prerequisites

  • Claude Code installed and working. See the Claude Code setup guide if you have not installed it yet.
  • A way to run Claude Code. You need one of: your own claude.ai subscription, Claude Code set up through CBorg , or Claude Code configured to run local models.
  • Internet access from wherever you run Claude Code, so it can fetch the documentation pages.

Install the plugin

This repository doubles as a Claude Code plugin marketplace, so you do not need to clone anything. From inside a Claude Code session, add the marketplace and install the plugin:

/plugin marketplace add lbnl-science-it/scienceit-docs
/plugin install lawrencium-hpc@scienceit-docs

Then reload plugins (or restart the session):

/reload-plugins

That's it. The skill loads automatically whenever your question matches its topic — you do not need to invoke it by name.

Using it

Just ask Claude Code your question in plain language. For example:

  • "Write me a Slurm script to run a 4-GPU PyTorch job on Einsteinium."
  • "How do I request a Lawrencium user account?"
  • "Which module do I load for Open MPI under GCC?"
  • "How do I move a large dataset onto the cluster?"

Claude recognizes that the question is about Lawrencium, fetches the relevant documentation page(s), and answers with a citation to the corresponding page on scienceit-docs.lbl.gov.

Allow fetching the docs without a prompt

The plugin fetches pages with Claude Code's WebFetch tool, so the first time it accesses a domain, Claude Code asks for your permission. To pre-approve the GitHub host that serves the raw Markdown, add a WebFetch rule to your settings. Use ~/.claude/settings.json to apply it everywhere, or a project's .claude/settings.json to scope it to that project:

{
  "permissions": {
    "allow": [
      "WebFetch(domain:raw.githubusercontent.com)"
    ]
  }
}

If a permissions.allow array already exists, add the string to it rather than duplicating the block. With this rule in place, fetching the documentation runs without prompting.

Other ways to ask

Prefer not to use Claude Code? See Asking LLMs for the Lawrencium AI assistant (lrc-agent), the CBorg Science IT assistant, and the single-file llms-full.txt context bundle you can drop into any large-context chat model.

Feedback

Questions or problems with the plugin? Contact the HPC Help Desk or email scienceithelp@lbl.gov.