Reproduce a Claude Code setup on any machine — any OS — by cloning this repo and talking to it. No install scripts: a running Claude reads this repo and applies the config for you.
- Claude Code installed and authenticated (the native binary is fine). A running Claude can't install or authenticate itself — do this once, manually.
- Node.js — needed only to render the status line (it's a Node script).
Native Claude doesn't bundle a general-purpose
node; install Node.js LTS to see the status line. Without it, everything still applies — the status line just won't draw until Node is on PATH.
git clone <repo-url> ai-setup
cd ai-setup
# open Claude Code with this folder as the working directory, then say:
# set me upClaude reads CLAUDE.md, shows a preview of exactly what it will write and where,
and applies it only after you approve. Existing files are backed up first.
- Aliases — global
claude<m>[<e>]shell shortcuts that launch Claude with a preset model + effort (e.g.claudeo= opus/max,claudeox= opus/xhigh); extra args forward through. 20 functions, installed into your shell profile. - Status line — a two-line bar (
config/statusline/statusline.js) wired up viasettings.json. Renders only whennodeis on PATH. - Settings — your
settings.jsonprefs (model, effort, theme, plugin enable-flags, …), deep-merged into your live config. - Plugins — installs the superpowers plugin (required) from the official marketplace; the settings flag only enables it, this step fetches it.
- Memory clean-start (opt-in) — disables Claude auto-memory in your home directory so home sessions start clean; you're asked first.
The set of things it configures grows over time; see CLAUDE.md for the live
manifest and docs/setup-guide.md for details, a self-test, and troubleshooting.
- OS-agnostic: machine-specific paths are resolved at setup time
(
${CLAUDE_HOME}→ your live config dir) and written with forward slashes. - Secrets are never committed. Credential handling (interactive prompts) is added when a payload first needs one.