Claude Code Environment Replication — Portable Setup Across Machines

A power user’s Claude Code setup is far more than the CLI binary — it’s 14 plugins, 236 skills, 11 custom commands, hooks, agents, templates, global instructions, and MCP server configs. Replicating that environment on a second machine (or recovering after a fresh install) is tedious if done manually. This entry documents the full ~/.claude/ anatomy, identifies what’s portable vs. machine-specific, and provides a one-command bootstrap script.

Quick Start — Replicate in 3 Steps

SOURCE MACHINE                    TARGET MACHINE
──────────────                    ──────────────
Step 1: Export                    Step 2: Bootstrap
  bash ~/.claude/                   bash claude-code-bootstrap.sh
    claude-code-export.sh
        │                         Step 3: Install plugins
        │  ~/claude-code-export/    (open Claude Code, paste
        └──── copy via USB/SCP ──►   plugin commands from
             AirDrop/Dropbox         bootstrap output)

Step 1 — Export (on source machine):

bash ~/.claude/claude-code-export.sh
# Creates ~/claude-code-export/ with 20 portable files (~132K)

Step 2 — Bootstrap (on target machine):

# Prerequisites: install Claude Code CLI first
npm install -g @anthropic-ai/claude-code

# Copy the export folder to the new machine, then:
bash claude-code-bootstrap.sh
# Installs config, fixes paths, checks dependencies

Step 3 — Install plugins (inside Claude Code on target machine):

# The bootstrap script prints these commands — paste them into Claude Code:
/plugin install frontend-design@claude-plugins-official
/plugin install superpowers@claude-plugins-official
# ... (14 total — full list printed by bootstrap)
/reload-plugins

Step 4 — Post-bootstrap auth:

codex login              # Authenticate Codex CLI
rtk gain                 # Verify RTK is working
# MCP servers (Calendar, Gmail, etc.) will prompt on first use

That’s it. Full environment replicated.

The Core Problem

Claude Code stores everything under ~/.claude/. Some of it is portable config you want on every machine. Most of it is runtime data you don’t. Knowing which is which saves hours of debugging “why doesn’t my new laptop have my skills.”

~/.claude/
├── PORTABLE (sync these)              ├── MACHINE-SPECIFIC (don't sync)
│   ├── CLAUDE.md                      │   ├── sessions/
│   ├── RTK.md                         │   ├── history.jsonl
│   ├── RESEARCH_WORKFLOW.md           │   ├── cache/
│   ├── settings.json                  │   ├── image-cache/
│   ├── commands/                      │   ├── paste-cache/
│   ├── agents/                        │   ├── file-history/
│   ├── hooks/                         │   ├── telemetry/
│   ├── templates/                     │   ├── usage-data/
│   └── (plugins reinstall from       │   ├── session-env/
│        marketplace names)            │   ├── shell-snapshots/
│                                      │   └── settings.local.json

What Lives in ~/.claude/ — Full Anatomy

The Three Replication Strategies

Strategy 1: Git Dotfiles Repo (Recommended)

The cleanest approach — version-control your portable config, symlink into ~/.claude/ on each machine.

~/dotfiles/claude/
├── CLAUDE.md
├── RTK.md
├── RESEARCH_WORKFLOW.md
├── settings.json
├── commands/
│   ├── editor.md
│   ├── mywiki.md
│   ├── reviewer2.md
│   └── ...
├── agents/
│   └── fast-reader.md
├── hooks/
│   └── rtk-rewrite.sh
├── templates/
│   └── research-CLAUDE.md
└── install.sh          ← bootstrap script

Pros: Version history, diff-able, works across macOS/Linux. Cons: Must remember to commit after changing config.

Strategy 2: Dropbox/iCloud Sync

Point ~/.claude/ symlinks at a synced folder. No git ceremony.

Pros: Automatic sync, zero friction. Cons: No version history, potential sync conflicts, syncs too much if you’re not careful.

Strategy 3: Export/Import Script

One-time export from source machine, one-time import on target. Best for occasional replication.

Pros: Simple, explicit control. Cons: Snapshot in time — doesn’t auto-update.

The Scripts

Two scripts handle the workflow: export (run on source machine) and bootstrap (run on target machine). Both live at ~/.claude/ and are included in the export bundle.

Export (Run on Source Machine)

bash ~/.claude/claude-code-export.sh [output-dir]
# Default output: ~/claude-code-export/

What it does:

1. Copies portable files: CLAUDE.md, settings.json, commands/, agents/, hooks/, templates/, memory/
2. Skips symlinks (e.g., /feynman -> ~/.codex/skills/feynman) with a warning
3. Guards against rm -rf on non-export directories
4. Prints a manifest of exported files

Bootstrap (Run on Target Machine)

bash claude-code-bootstrap.sh [path-to-exported-config]
# Default: auto-detects from script's own directory

What it does:

1. Copies all config files into ~/.claude/
2. Backs up existing settings.json before overwriting
3. Rewrites absolute paths (e.g., /Users/oreo -> $HOME) using jq (with sed fallback)
4. Prints plugin install commands to run inside Claude Code
5. Checks all external dependencies and reports missing ones
6. Prints a post-bootstrap checklist

Plugin Install Commands (Run Inside Claude Code)

After bootstrap, open Claude Code and run:

# Official marketplace (8 plugins)
/plugin install frontend-design@claude-plugins-official
/plugin install playground@claude-plugins-official
/plugin install pr-review-toolkit@claude-plugins-official
/plugin install claude-code-setup@claude-plugins-official
/plugin install superpowers@claude-plugins-official
/plugin install playwright@claude-plugins-official
/plugin install github@claude-plugins-official
/plugin install claude-md-management@claude-plugins-official

# Custom marketplaces — add sources first
/plugin marketplace add jimliu/baoyu-skills
/plugin marketplace add kepano/obsidian-skills
/plugin marketplace add Lum1104/Understand-Anything
/plugin marketplace add lingfengQAQ/webnovel-writer
/plugin marketplace add openai/codex-plugin-cc

# Then install from custom marketplaces (6 plugins)
/plugin install utility-skills@baoyu-skills
/plugin install ai-generation-skills@baoyu-skills
/plugin install obsidian@obsidian-skills
/plugin install understand-anything@understand-anything
/plugin install webnovel-writer@webnovel-writer-marketplace
/plugin install codex@openai-codex

/reload-plugins

The full script source lives at ~/.claude/claude-code-export.sh and ~/.claude/claude-code-bootstrap.sh — they’re included in every export bundle so the target machine is self-contained.

What’s in Q’s Actual Setup (Reference Inventory)

14 Plugins

11 Custom Commands

2 Hooks

1 Custom Agent

1 Template

External Dependencies

Gotchas and Tips

Paths in settings.json

The RTK hook path in settings.json is absolute (e.g., /Users/oreo/.claude/hooks/rtk-rewrite.sh). The bootstrap script automatically rewrites these paths to match your new $HOME using jq (or sed as fallback). If you skip the bootstrap and copy settings.json manually, you’ll need to fix paths yourself:

# Only needed if you skip the bootstrap script:
jq --arg old "/Users/oreo" --arg new "$HOME" \
  'walk(if type == "string" then gsub($old; $new) else . end)' \
  ~/.claude/settings.json > /tmp/settings.json && mv /tmp/settings.json ~/.claude/settings.json

Plugin Versions Drift

Plugins auto-update from their marketplace. After bootstrapping, your new machine gets the latest versions — which may differ from the source. This is usually fine, but if a skill breaks, check the plugin version with /plugin list.

MCP Servers Need Re-Auth

MCP servers (Google Calendar, Gmail, Google Drive, alphaxiv) store auth tokens locally. You must re-authenticate on the new machine. Claude Code will prompt you automatically on first use.

Feynman Symlink

The /feynman command is a symlink to ~/.codex/skills/feynman. If Codex CLI isn’t installed yet, this symlink will be broken. Install Codex first, then the symlink resolves.

Memory Files Are Optional but Valuable

The projects/ directory (1.6 GB) contains per-project memory that Claude uses for context across sessions. You can skip syncing it — Claude will rebuild context over time — but copying key project memories accelerates onboarding on the new machine.

How LearnAI Team Could Use This

• Lab machine setup — New students or TAs can bootstrap a full Claude Code environment in minutes instead of discovering features over weeks
• Consistent research tooling — Every team member gets the same commands (/reviewer2, /lit-scout, /editor), hooks (RTK token savings), and review gates (Codex)
• Teaching workshops — Export a minimal config (CLAUDE.md + 3-4 key commands) as a “starter kit” for students learning Claude Code
• Disaster recovery — If a machine dies or gets reimaged, the export file + 5 minutes of plugin installs gets you back to full capability

Real-World Use Cases