15 Claude Code tips that save hours every week

Short answer: the biggest wins with Claude Code come from three habits: give it a short CLAUDE.md with your project's rules, ask for a plan before any code, and keep each session to one task you can review as a single diff. The other twelve tips below build on those three.

Claude Code is Anthropic's coding agent. It reads your repository, runs commands and edits files across the codebase. It is powerful out of the box, but a few habits decide whether it saves you an afternoon or costs you one.

Disclosure: we build Zokie, an agent IDE that runs Claude Code. The tips below work with or without it.

Set it up once

1. Write a CLAUDE.md before your first real task

Claude Code reads a CLAUDE.md file in your project at the start of every session. Put the things you would tell a new teammate in it: how to run the tests, which folders matter, naming rules and what never to touch. Keep it under a page. Our guide to CLAUDE.md and AGENTS.md has a template.

2. Let /init draft it for you

Running /init in a new project asks Claude Code to read the repository and write a first CLAUDE.md. Treat the result as a draft: delete anything generic and add the rules only your team knows.

3. Decide what it may run without asking

Claude Code asks before it runs commands or edits files. Allow the safe, frequent ones (your test runner, your linter) and keep asking for anything that deploys, deletes or touches production. You will approve less and worry less.

Work in small, reviewable tasks

4. One task per session

Long sessions drift. When a session starts mixing a bug fix with a refactor and a new feature, the diff becomes impossible to review. Finish one task, review it, then start a fresh session for the next.

5. Ask for a plan first

For anything bigger than a one-line fix, ask: "Read the code and propose a plan. Do not edit yet." Plan mode does the same thing. Correcting a plan takes a minute; correcting a finished change takes an hour.

6. Name the files and the done condition

"Fix login" is vague. "In auth/session.ts, expired tokens should return 401, not 500; the test in session.test.ts should pass" is a task an agent can finish and you can check. Our guide to writing tasks for coding agents goes deeper.

7. Make it run the tests

Agents are much better when they can check their own work. Tell Claude Code which command runs the tests and ask it to run them before it says it is done.

Use the bigger features

8. Let sub-agents take the side quests

Claude Code can hand parts of a task to sub-agents, such as searching the codebase or reviewing a change, so the main session keeps a clean context. Read sub-agents explained for when they help and when they just cost tokens.

9. Add a few MCP servers, not twenty

MCP servers connect Claude Code to tools such as your issue tracker, database or docs. Each one adds context and options. Start with the one or two you use daily; see MCP servers explained.

10. Use hooks for rules that must never be broken

A rule in CLAUDE.md is a request. A hook is a script that runs on events such as a file edit, so it can enforce things: run the formatter after every edit, or block edits to a generated folder.

11. Save repeated prompts as custom commands

If you type the same instructions every week ("review this diff for security issues", "write release notes from the last ten commits"), save them as custom slash commands so the whole team uses the same wording.

Review like it matters

12. Read the diff, not the chat

The agent's summary tells you what it meant to do. The diff tells you what it did. Always review the diff. Our 10-point checklist for reviewing AI-written code is a good routine.

13. Run parallel tasks on separate branches

Two sessions editing the same folder will overwrite each other. Give each task its own git worktree and branch. Git worktrees explained shows how, by hand or with an agent IDE.

14. Match the model to the task

A version bump does not need the biggest model. Pick a smaller model for routine work and save the frontier model for design and hard bugs. See small vs frontier models.

15. Keep the terminal, add a window when you outgrow it

The terminal is the most flexible way to use Claude Code. Once you run several sessions at once, a desktop app helps you see them all. Our comparison of desktop apps for Claude Code covers the options, including Zokie, where every task gets its own worktree and ends as a diff you approve.

FAQ

What is the most useful Claude Code tip for beginners?

Write a short CLAUDE.md with how to run and test your project, and ask for a plan before any edits. Those two habits prevent most bad changes.

How do I stop Claude Code from asking permission for every command?

Allow the safe commands you run all the time, such as your test runner and linter, and keep approval on for anything that deploys, deletes data or touches production.

Can I run several Claude Code sessions at once?

Yes, as long as each session works in its own git worktree and branch so they do not overwrite each other's files. Agent IDEs such as Zokie set this up for every task automatically.

Does Claude Code work on Windows?

Yes. Claude Code runs on macOS, Linux and Windows, and Zokie runs it on macOS and on Windows 10 and 11.