onboard

name	onboard
description	Onboard a lab member onto the shared Claude Code and/or Codex configuration. Detects clusters, collects Slurm account details, runs setup, and helps customize personal config.

Lab AI Coding Setup - Onboarding

Help a new lab member set up the shared lab configuration for Claude Code, Codex, or both. Be friendly, concise, and beginner-friendly.

Start with a greeting like: "Welcome onboard, Foreseer!" Then briefly explain that this setup teaches the coding assistant about the lab's cluster environment, GPU accounts, storage rules, and Slurm workflows.

Pre-flight Checks

1. Check whether the user has Claude Code and/or Codex initialized:

ls -ld ~/.claude ~/.codex 2>/dev/null

• If neither exists, ask them to run claude or codex once first, depending on what they want to use.
• If both exist, default to configuring both.
• If only one exists, configure that one unless the user asks otherwise.

2. Check if the lab config repo is cloned at ~/lab-claude-config/. If not, tell the user to clone it:

git clone https://github.com/umich-foreseer/lab-claude-config.git ~/lab-claude-config

3. Verify lightweight dependencies:

which jq 2>/dev/null

jq is needed for Claude's statusline. If the user is configuring only Codex, missing jq is not blocking.

Detect Cluster and Accounts

Do this proactively. Run commands yourself and present what you found.

1. Detect username

whoami

2. Detect cluster

hostname -f
sinfo -o "%12P %16G" --noheader 2>/dev/null | head -20

• If spgpu2 exists, include Great Lakes.
• If hostname contains lighthouse or lh-login, include Lighthouse.
• If unclear, ask the user.

3. Look up Slurm accounts

sacctmgr show association user=$(whoami) format=account%20,partition%20,qos%40 --noheader 2>/dev/null

For an owned account, also check the group memory cap:

sacctmgr show association account=<owned_account> format=account%20,grptres%40 --noheader 2>/dev/null | head -5

Use this reference table:

Account	Cluster	Partition(s)	GPUs	Billing
qdj_project_owned1	Great Lakes	spgpu2	L40S	Prepaid/shared dynamic quota
qmei0	Great Lakes	gpu, gpu-rtx6000, standard	V100, RTX Pro 6000 Blackwell	Prepaid
qmei3	Great Lakes	gpu, gpu-rtx6000, standard	V100, RTX Pro 6000 Blackwell	Pay-as-you-go
qmei	Lighthouse	qmei-a100	A100	Dedicated

Only present accounts for the detected cluster.

Confirm Findings

Show a short summary with username, cluster, accounts, partitions, and any memory cap. Ask whether it looks right before writing config values.

Run Setup

Write ~/lab-claude-config/build/.env.local with only the variables for the detected cluster.

For Great Lakes:

# Lab Claude Config - saved template variables
GL_USERNAME=<value>
GL_ACCOUNT_OWNED=<value>
GL_ACCOUNT_GENERAL=<value>
GL_MEMORY_CAP=<value>

For Lighthouse:

# Lab Claude Config - saved template variables
LH_USERNAME=<value>
LH_ACCOUNT=<value>
LH_PARTITION=<value>
LH_GPU_COUNT=<value>
LH_GPU_TYPE=<value>

Then run setup from the repo:

cd ~/lab-claude-config && ./setup.sh --modules <modules> --targets <targets> --non-interactive

Examples:

# Codex only
./setup.sh --modules greatlakes --targets codex --non-interactive

# Claude Code and Codex
./setup.sh --modules greatlakes,lighthouse --targets claude,codex --non-interactive

Post-setup

Summarize what was installed:

• Claude: lab block in ~/.claude/CLAUDE.md, settings, hooks, skills, and agents.
• Codex: lab block in ~/.codex/AGENTS.md and skills in ~/.codex/skills.

Ask whether they want to add personal notes outside the lab markers. Good examples:

• Their turbo storage path.
• Current project paths.
• Language/framework preferences.
• Personal coding conventions.

Remind them:

• To update: cd ~/lab-claude-config && git pull && ./setup.sh --targets <targets>
• Personal content outside the lab markers is preserved.
• In Codex, ask for skills by name, for example "use the slurm-status skill".

If Setup Fails

Read the error output and help debug. Common issues are missing jq for Claude, wrong account names, or a missing ~/.claude / ~/.codex directory.