The Setup
From "I've heard about Claude Code" to a terminal that auto-launches a working studio in one keystroke.
8 modules · 23 lessons · ~3.5h
Get a Claude Pro/Max sub. What you get vs. the API. Estimated cost for a hobbyist vs. a working studio.
- Get the sub. The Pro/Max math.
What a terminal is, install Ghostty, read our config.
- What a terminal is, in 4 minutes
- Install Ghostty
- Our Catppuccin Mocha config — every line, why
Why Zellij, install, the Alt-keybinding muscle memory, detach + reattach.
- Why Zellij (vs. tmux, vs. nothing)
- Install + your first session
- The Alt-keybindings — Alt+h/j/k/l, splits, tabs, detach
- Detach and reattach — sessions outlive windows
Install, settings.json, the autonomy ladder.
- npm i -g @anthropic-ai/claude-code, first launch
- ~/.claude/settings.json — permissions, plugins, voice
- The autonomy ladder: cautious → auto → dangerous
macOS desktop split, single-pane vs 2×2, Yazi as a pane.
- Finder pinned left, Ghostty pinned right
- Two layouts — single-pane and the 2×2
- Yazi as a pane — file browser without leaving
dev / dev-new / dev-done — read every line of the bash.
- dev — FZF picker → Zellij → Claude auto-launches
- dev-new — branch + worktree at {project}@{branch}
- dev-done — kill, merge, prune, gone
- Read every line of the bash. You'll customize all of it.
The studio-inventory pattern + per-project memory.
- Global CLAUDE.md — the studio inventory pattern
- Per-project CLAUDE.md — six lines that matter
Plan mode, AskUserQuestion, scope discipline, ExitPlanMode.
- Plan mode — the kickoff prompt
- AskUserQuestion + scope discipline
- ExitPlanMode and the approval ritual