Claude Code Starter Checklist

A checklist for getting Claude Code working on a real codebase. The goal is to spend less time fighting defaults and more time shipping.

Before your first prompt

• Write a CLAUDE.md at the repo root. Tech stack, commands (install, dev, test, format), conventions, voice rules for generated content.
• Check in a .claude/settings.json with repo-wide permissions so teammates share the same allowlist.
• Run /init once to let Claude Code generate a first-pass CLAUDE.md, then edit it down. The auto-generated one is always too long.

Ground rules worth enforcing

• UK vs US spelling. Pick one, put it in CLAUDE.md, stick to it.
• No emojis in code or content unless you want them. The default behaviour sprays them everywhere.
• No trailing summaries. If the diff is visible, a "here's what I changed" paragraph is noise.
• One concern per commit. Easier to review, easier to revert.

When Claude gets stuck

• Ask it to run the failing command itself and read the output before guessing.
• If it's hallucinated a library function, point at the real docs — do not let it keep guessing.
• For ambiguous tasks, ask for a 2-3 sentence recommendation before any implementation.

Things not to bother with

• Multi-thousand-line CLAUDE.md files. Nobody reads them, including the model.
• Custom slash commands for one-off tasks. Use them for things you do weekly, not once.
• Hooks that duplicate what your linter already does. Let Pint, ESLint, etc. do their job.

Further reading

• docs/BUILD-BRIEF.md in any repo that uses this pattern
• samjenkins.com/writing — working examples of the above