Part 1
Install and sign in
Study first
- Claude Code: the memory page, which explains
CLAUDE.mdandAGENTS.md, and the permission modes. You should be able to say what the default mode asks before, each shell command and each file write, and whatacceptEditschanges. - Ghostty: tabs, splits and the configuration file. From Day 4 you run several agents at once, one per split.
Build
- Install Ghostty, Claude Code and Cursor. Sign in to Claude Code, sign in to GitHub with
gh auth login, and create a Vercel account linked to the same GitHub account. In Claude Code, add the Codex plugin with/plugin marketplace add openai/codex-plugin-ccand/plugin install codex@openai-codex, then run/codex:setup; it needs a ChatGPT account or an OpenAI API key. - On the setup page, create your training repository from the template and enter its name, so the course can read it when you verify a day. Clone it.
- Open the clone in a Ghostty tab with two splits, Claude Code in one and a shell in the other. Open the same folder in Cursor with the ESLint and Prettier extensions installed, so the editor shows the problems the checks will report.
Done when
claudestarts inside the clone,gh auth statusnames the right account, and Cursor flags an unused variable in a file you open.
Technical notes
- Claude Code loads
CLAUDE.md, orAGENTS.mdin its place, at the start of every session, from the working directory up to the repository root. One file is enough. The reference keeps the conventions inAGENTS.md, which Codex reads too, and aCLAUDE.mdcontaining only@AGENTS.md. - In the default permission mode, Claude Code asks before running a shell command or writing a file; reading files does not ask.
acceptEditsstops the prompts for file edits and keeps them for commands. Switch oncepnpm checkexists, since the check then catches a bad edit instead of the prompt. - Ghostty reads
~/.config/ghostty/config. Two keybinds, new split right and new split down, cover the course.


