lab-skills

Lab Claude Skills

Shared Claude Code skills and conventions for the lab. These skills teach Claude Code our lab's standards for data handling, plotting, script organization, reproducibility, and more.

What are skills?

Claude Code skills are markdown files that give Claude specific instructions for particular tasks. They're how we teach Claude our lab's conventions.

How they work: Each skill has a short description that tells Claude when to activate it, and a body with the actual instructions. Some skills also include additional files in their folder — guides, templates, or reference material that Claude can use.

There are two kinds:

• Automatic skills load on their own when Claude detects a relevant task. For example, when you ask Claude to make a plot, r-plotting-style loads automatically and Claude follows our ggplot2 conventions. You don't need to do anything.
• User-invoked skills (slash commands) are triggered by typing a command like /done or /new-project. These run specific workflows on demand.

These skills are under active development. If Claude does something wrong, or you think a skill is missing guidance for a situation you ran into, or you have an idea for a new skill — open a GitHub issue. See Improving skills below for details.

Prerequisites

You need Claude Code installed and working before adding lab skills:

1. Install Claude Code — see our lab compute handbook or Anthropic's install guide
2. Open Claude Code — either in Positron (click the Claude Code icon in the sidebar) or in your terminal (run claude)

Install

There are two ways to install. Choose whichever fits your needs.

Option A: Plugin (recommended)

The simplest way to get all lab skills and keep them up to date.

In Positron / VS Code:

1. Type /plugins in the Claude Code chat panel to open the plugin manager
2. Go to the Marketplaces tab
3. Add MusserLab/lab-claude-skills
4. Switch to the Plugins tab and install lab-skills

In the terminal CLI:

/plugin marketplace add MusserLab/lab-claude-skills
/plugin install lab-skills

This installs all skills and hooks as a single package. Skills are available as /lab-skills:skill-name (e.g., /lab-skills:done). Security hooks activate automatically — see Security below.

Updates:

If you installed via the marketplace (Option A), the plugin updates automatically when you restart Claude Code. No manual action needed — just restart when prompted.

If auto-update doesn't work or you installed via CLI, you can manually update by opening a terminal, running claude, then:

/plugin uninstall lab-skills
/plugin install lab-skills

Option B: Manual install (if you want to customize)

If you want to modify skills or pick only the ones you need, clone the repo and copy skills into your personal Claude Code directory instead.

# Clone the repo
git clone https://github.com/MusserLab/lab-claude-skills.git
cd lab-claude-skills

# Copy all skills to your personal skills directory
cp -r skills/* ~/.claude/skills/

# Or copy just the ones you want
cp -r skills/data-handling ~/.claude/skills/
cp -r skills/r-plotting-style ~/.claude/skills/

To update, pull the repo and re-copy the skills you want:

cd lab-claude-skills && git pull
cp -r skills/data-handling ~/.claude/skills/   # update specific skills

Skills that reference machine-specific paths (like conda-env) use ~/miniconda3 as the default. If your conda is installed elsewhere, edit your local copy.

Setup

After installing, complete these steps to configure permissions and security protections.

If you used the plugin install and don't have the repo cloned, download template files from templates/ on GitHub.

1. Copy the settings template

On your local machine (macOS/Linux desktop):

cp templates/settings-example.json ~/.claude/settings.json

On an HPC cluster (Bouchet, McCleary, etc.):

cp templates/settings-cluster.json ~/.claude/settings.json

The cluster template includes glob-based deny rules for Linux, blanket Bash(*) permissions (hooks handle safety), extra protections for shared filesystems (rm -rf /nfs*, process killing), and disables telemetry. It also has a YCRC cluster quick reference in the comments.

If you already have a settings.json, merge in the entries you want manually.