doyouknow.ai
Sign in

How to Use Claude Code: Your First Week

By , AI Implementation Coach · · 11 min read

Banner: Your first Claude Code task should be a question. Ask first, then plan, then ship. One habit a day, all week.

Your first Claude Code task should be a question.

You installed Claude Code, the cursor blinks, and you don't know what to ask. Ask about your code first. This first-week plan shows you how to use Claude Code one habit a day, all taken from Anthropic's own guides.

If you only read this

  • Day 1: ask questions. Don't ask for code yet.
  • Keep CLAUDE.md short: only what Claude can't find in the repo.
  • Plan before big changes, and give Claude a test to run.
  • Corrected it twice and still wrong? /clear and write one better prompt.

Below: what Claude Code does when you give it a task, one habit for each day of your first week, and what to add in week two. On day 2 you'll also find the one line in CLAUDE.md that saves the most time.

Getting started with Claude Code is easier on a project you know. The examples here use a made-up one: a small web app with a login page and a set of tests. Swap in your own.

Rolling Claude Code out to a team? Message me on WhatsApp and tell me where you're stuck.

Claude Code reads, acts and checks, and you can stop it any time

Claude Code is a coding agent. It works in your terminal, on your project. You give it a task in plain words, and it works through three steps: gather context, take action, check the result. It repeats them until the task is done.

You stay in that loop. Press Esc and it stops right away, keeps what it has done so far, and waits for you. Before it edits a file, it saves a copy, so you can roll the change back.

If you haven't installed it yet, Anthropic's quickstart has the command for your system. Then open a terminal in your project folder and type claude.

Day 1: Ask it about your code

Don't ask for code on day 1. Ask questions. Anthropic's quickstart starts the same way:

what does this project do?

Then go narrower:

where is the main entry point?
explain the folder structure
how does login work in this app?

You don't need to paste files in. Claude Code reads your project files as needed. Watch which files it opens. That tells you how it finds its way around your code.

Why start here? Because you learn how it thinks before you trust it with a change. And you'll spot what it gets wrong about your project, which is exactly what goes into tomorrow's file.

Day 2: Write a short CLAUDE.md

CLAUDE.md is Claude Code memory for your project. It's a plain text file in your repo, and Claude reads it at the start of every session.

Run /init. It looks at your code and writes a first CLAUDE.md with build commands, test steps and the conventions it found. Then cut it down.

Keep it short, because every line loads into every session, whether the task needs it or not. Anthropic's Claude Code best practices say: write down only what Claude can't find by reading the code. Keep these:

  • How to check work: the exact test, typecheck and lint commands, and how to run one test file.
  • Odd corners: an old pattern new code must not copy, or a folder that looks unused but isn't.
  • Decisions that aren't in the repo: how you deploy, which helper the team settled on.
  • Anything you've corrected twice.

The one line that saves the most time is your test command. Without it, Claude stops when the work looks done. With it, Claude can check its own work as it goes, and you stop being the test suite.

A short one looks like this:

Commands
- Run all tests: npm test
- Run one test file: npm test -- path/to/file.test.js

Gotchas
- src/legacy/ is still live. Don't copy its patterns.

Already have an AGENTS.md file for other coding agents? Claude Code reads it when there's no CLAUDE.md, and one setting makes it read both. Claude also keeps its own notes from your corrections (auto memory). Use /memory to see and edit both.

Day 3: Plan before you change anything

On day 3, try Claude Code plan mode. In plan mode, Claude reads your files and proposes a plan, but makes no edits until you approve it.

Press Shift+Tab until the status bar shows plan mode, or start one prompt with /plan:

/plan add a "remember me" option to the login page

Read the plan. Fix it if it's wrong: tell Claude, or press Ctrl+G to edit the plan yourself. Then approve it, and Claude starts editing.

Use plan mode for anything that touches more than one file. For a typo, skip it.

Not sure what you want yet? Ask Claude to interview you first: "Interview me until you could write the spec, then write it to SPEC.md." Then start a fresh session and build from that file.

Day 4: Ship one small fix, with a test to run

On day 4, ship something real, but small. Pick a bug you already understand.

  1. Give it the symptom, the place and what "fixed" means. Anthropic's own example: instead of "fix the login bug", write "users report login fails after the session times out. look at token refresh in src/auth/, write a failing test that reproduces it, then fix it" (best practices).
  2. Let it reproduce the bug with a failing test. Now there's a check it has to pass.
  3. Ask it to run the tests and show you the output. Anthropic says to ask for evidence: the command it ran and what it printed. "All tests pass" on its own tells you little.
  4. Read the diff. Run /diff to see every change it made.
  5. Open a pull request. Type "create a pr". Review it before you submit it, and ask Claude to point out the risks.

Want this set up properly for your team? I help teams set up coding agents on their own projects: the instructions file, the checks and the review steps. Chat on WhatsApp, or see implementation.

Day 5: Stay in control

By day 5 you know what Claude Code does. Now choose how much it may do without asking. That's the permission mode, and Shift+Tab switches between them:

  • Manual: it asks before every edit and every command that isn't read-only.
  • Accept edits: it edits files without asking, and still asks before other commands.
  • Plan: it reads and proposes, and edits nothing until you approve.
  • Auto: it works without stopping for routine approvals, and a separate safety check blocks risky actions. In the terminal, new sessions now start in this mode unless you change it. Want to see every step in week one? Switch to Manual.

There's also a flag that skips every permission check. Anthropic says to use it only in a container or a virtual machine, where Claude can't harm your own computer. Never on your laptop.

When it goes the wrong way, you have three ways back:

The Claude Code commands worth knowing in week one (full list in the commands reference):

Command or key What it does
/init Writes a first CLAUDE.md from your code
/memory Opens your CLAUDE.md files and Claude's own notes
/plan Plan mode for one prompt
Shift+Tab Switches the permission mode
Esc Stops Claude, keeps the work so far
/rewind Rolls code or chat back to an earlier point
/diff Shows every change in your working tree
/clear Starts a fresh conversation
/compact Shortens a long conversation into a summary
/permissions Shows and edits what Claude may do without asking
/usage Shows what your sessions are using
/help Lists the commands you have

Week two: add skills, hooks and helpers only when you need them

Don't set everything up on day 1. Anthropic's guide lists the sign that tells you it's time for each one:

  • Skills: you keep typing the same prompt to start a task. Save it as a skill and run it with one command. Claude Code skills can also load on their own when a task needs them.
  • Hooks: you want something to happen every time, like a check after each edit. That's a hook, and unlike a line in CLAUDE.md, it always runs.
  • Subagents: a side task fills your chat with output you won't read again. Send it to a subagent, which returns only a summary.
  • MCP: you keep copying data from a browser tab Claude can't see. Connect that tool with MCP.

One trigger comes first: if Claude gets the same thing wrong twice, add a line to CLAUDE.md.

Keep the bill down from day one

Every request sends the whole conversation again. So a session that has collected three unrelated tasks re-reads all three on every turn of the fourth. Three habits keep that in check:

  • /clear between unrelated tasks. It costs nothing.
  • Keep CLAUDE.md short. Every line is read on every task.
  • Check /usage. On paid plans it shows what is using up your plan, by skill, subagent and tool.

For where the rest of a coding agent bill goes, and how teams cut it without a worse model, read my post on why coding agent costs keep growing. Plans change often, so check Anthropic's pricing page for what each one includes today.

My take: treat it like a new teammate

You wouldn't give a new developer a big feature on their first morning. You'd show them the code, tell them how you test, and give them a small bug. Claude Code is the same.

So: questions first. Then a short CLAUDE.md. Then small tasks with a test to pass. Real features come once it knows your project and you know how it works.

Who should follow this plan: any developer or small team starting with Claude Code on a real codebase. Who can skip parts: if you already use another coding agent every day, jump to day 2 and day 5. The instructions file and the permission modes are where Claude Code differs most.

New to AI tools in general? Start with my guide on how to learn AI.

Want help with Claude Code on your team?

If your team is starting with Claude Code, or already uses it and the results are uneven, tell me what you're working on. I'll tell you where I'd start.

Training a developer team? See AI coaching and training.

Chat on WhatsApp

For a done-for-you setup on your own projects, see implementation.

FAQ

Is Claude Code free?

No. Claude Code needs a paid Claude plan or an API account in the Claude Console. The free Claude plan doesn't include it. Plans and what they include change often, so check Anthropic's pricing page.

What should I put in CLAUDE.md?

Only what Claude can't find by reading your code: your test, typecheck and lint commands, parts of the code that mislead, team decisions that aren't written down, and anything you've corrected twice. Start with /init, then cut. Anthropic's guide has a short example.

What is plan mode in Claude Code?

Plan mode makes Claude read your files and propose a plan without editing anything. You approve, change or reject the plan before any code changes. Press Shift+Tab to switch to it, or start a prompt with /plan (permission modes).

Can I use Claude Code in VS Code?

Yes. There's a Claude Code extension for VS Code with a chat panel, changes shown in the editor and plan review. The habits in this post work the same there.

Do I need to know how to code to use Claude Code?

Not to start. Anthropic has a terminal guide for first-time users. But you should be able to read a change before you keep it. If you can't yet, use day 1's questions to learn your project first, and don't let it change code you can't check.

Khalil Kothia, AI Implementation Coach

Written by

Khalil Kothia

AI Implementation Coach

Khalil Kothia is an AI Implementation Coach who has spent 20 years building technology and now helps businesses put AI to work.

More about Khalil · LinkedIn

← All posts