How to Use Claude Code, Start to First Commit

By the end of this guide you will have Claude Code installed, signed in, and one small change in your own project reviewed and committed. The shortest path is five commands and three prompts: install, cd into a repo, run claude, ask what the project does, ask for one small change, approve it, and ask for a commit. Allow about twenty minutes. Facts checked 31 October 2026.

What you will have at the end

• Claude Code running in your terminal, on a plan you chose on purpose
• A first session where it explained a codebase and made one change
• A git commit you read before it was made
• A short list of commands to steer, stop and undo a session
• A clear idea of what it costs and what leaves your machine

This guide is the first-run guide. If you want to build a whole small app with it, go to the Claude Code tutorial. If you already use it daily and want rules for a team, read Claude Code best practices.

Prerequisites

You need three things.

1. A terminal. On macOS that is Terminal or iTerm. On Windows, PowerShell works. On Linux, any shell. The docs link a terminal guide if you have never opened one.
2. A project with git. Pick a real repo, but one where a bad edit would not hurt. A side project is ideal. Run git status first and make sure the working tree is clean, so the diff you see later is only Claude's.
3. A way to pay. Claude Code accepts a Claude subscription (Pro, Max, Ultra, Team or Enterprise), a Claude Console account with pre-paid credits, or access through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. See Cost below before you pick.

On native Windows the docs recommend installing Git for Windows, so Claude Code can use the Bash tool. Without it, Claude Code uses PowerShell as its shell tool. WSL setups do not need it.

Step 1: Install

The native installer is the recommended route. It also keeps itself up to date in the background.

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell
irm https://claude.ai/install.ps1 | iex

:: Windows CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

The installer shows no progress while it downloads, so do not assume it hung. Two other routes exist if you prefer a package manager:

# Homebrew (stable channel, about a week behind latest)
brew install --cask claude-code

# WinGet
winget install Anthropic.ClaudeCode

Homebrew and WinGet installs do not update themselves. You run brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode when you want a new version. There is also a claude-code@latest Homebrew cask that gets releases as soon as they ship. Debian, Fedora, RHEL and Alpine users can use apt, dnf or apk packages, listed on the setup page.

Open a new terminal window and check:

claude --version

A working install prints a version number followed by (Claude Code). If your shell says the command is not found, the install folder is not on your PATH yet. The docs have a "Fix your PATH" section for it.

Step 2: Start a session and sign in

Go to your project and start Claude Code.

cd /path/to/your/project
claude

On first use it asks you to log in, and finishes the sign-in in your browser. If you have ANTHROPIC_API_KEY set and you approve the key when asked, it skips the login prompt and bills your Console account instead. To switch accounts later, type /login inside the session.

You will see the version, the current model and the working directory above the prompt. Two commands to remember straight away:

• /help lists the commands.
• /resume continues an earlier conversation.

Step 3: Ask questions before you ask for changes

Do not start with a change. Let it read first. Claude Code opens the files it needs on its own, so you do not paste anything.

what does this project do?

where is the main entry point?

explain the folder structure

This does two jobs. It shows you whether Claude Code understood the repo, and it warms up the session with the right context. If an answer is wrong, you have learned something cheap: the repo is confusing, or the question was vague.

You can point at a specific file with @ and a path, which includes the file's contents in your message. You can also paste a screenshot of an error, or pipe output in from your shell:

cat error.log | claude -p "what is going on here"

The -p flag runs one prompt and exits. It is how you use Claude Code in scripts.

Step 4: Make one small change

Pick something you could do yourself in five minutes. A good first change has a clear finish line.

add input validation to the signup form so an empty email shows an error. Follow the pattern used by the other forms.

Compare that with "improve the forms". The first names the symptom, the place and the style to follow. The second makes Claude guess. The best-practices page puts it this way: give it information it cannot get from reading the repo, and the specific prompt saves you a round of corrections.

Claude Code shows you the proposed edit. In the default mode it asks before it writes. Read the diff and choose Yes only when you would accept the same change from a colleague.

What the permission modes do

Press Shift+Tab at any time to cycle through modes. These are the ones in the permissions docs:

The best-practices page says new terminal and VS Code sessions start in auto mode. If you would rather see every step on day one, start in Manual:

claude --permission-mode manual

For a bigger change, try plan mode first. Claude reads files and proposes a plan, and nothing is written until you approve it.

claude --permission-mode plan

Step 5: Run the checks

A change is not done when Claude says it is done. Give it something that returns pass or fail. If your project has tests, say so:

run the tests for the signup form and keep going until they pass

If you have a linter or a type check, name that too. The docs make the same point from the other direction: when Claude has nothing to run, "looks done" is all it has to go on, and you become the test suite. Ask for evidence, meaning the command it ran and its output, not just "all tests pass".

Step 6: Review and commit

Ask what changed, then look yourself.

what files have I changed?

git diff

Then let Claude write the commit:

commit my changes with a descriptive message

Claude Code handles branches too: create a new branch called feature/signup-validation, show me the last 5 commits, and help me resolve merge conflicts all work as plain requests. Check the message with git log -1 before you push anything. If your team has a rule about commit trailers or attribution, say it in the prompt or in a CLAUDE.md file, because Claude Code has its own defaults for commit text.

Step 7: Steer, stop and undo

These controls matter more than any prompt trick.

• Esc stops Claude mid-action and keeps what it did so far.
• Esc twice on an empty prompt, or /rewind, opens a menu to roll back code, conversation or both.
• /clear starts a fresh conversation. After two failed corrections, the docs advise /clear and one better prompt instead of a third correction.
• /compact summarizes a long session. Add a focus, such as /compact keep the API changes and the test commands.
• claude -c continues the most recent conversation in the current folder. claude -r lets you pick an older one.

Rewind does not cover everything. Checkpoints track files Claude edited with its own file tools. A file changed only by a shell command is not restored. Git is the real safety net, which is why we asked for a clean tree in the prerequisites.

Step 8: Teach it your project once

You will soon notice you repeat yourself: how to run the tests, which folder holds what. Put that in a file called CLAUDE.md at the project root. Run /init and Claude writes a first version from your code. The memory docs suggest staying under 200 lines and writing instructions you can verify, such as "Run npm test before committing" rather than "test your changes".

Your own preferences go in CLAUDE.local.md, which you add to .gitignore. For the deeper version of this, with permissions, hooks and subagents, see Claude Code best practices and our CLAUDE.md guide.

Where to use it other than the terminal

The engine is the same in every place, so settings carry over. The docs list the web (claude.ai/code), a desktop app, VS Code, JetBrains IDEs, Slack, and CI through GitHub Actions and GitLab. A terminal is the best place to learn it, because you see every step. After that, our Claude Code overview covers each surface.

What can go wrong

These come from the install, permissions and costs docs, not from our own testing.

Shell access is the real risk. Claude Code can run commands, so treat a session like a contractor with your keyboard. Keep unfamiliar code in Manual mode, never run bypassPermissions on a machine you care about, and read the permissions docs before you add allow rules.

What it costs

Claude Code is not sold alone. It comes with a Claude plan, with API billing, or through a cloud provider. The consumer plans from the pricing page, checked 31 October 2026:

Plans are subject to change at Anthropic's discretion, and prices exclude tax. Usage limits reset on a five-hour window plus a weekly cap, and how fast you reach them depends on the model, the length of your chats and the size of your files.

If you pay by API token instead, the costs page says enterprise deployments average about $13 per developer per active day, or $150 to $250 per developer per month, with costs under $30 per active day for 90% of users. Treat that as a planning figure for teams, not a quote for you. Run /usage in a session to see where your own usage goes. For more on plans, see Claude pricing and Claude Code pricing.

Two habits cut cost without cutting quality: /clear between unrelated tasks, and a smaller model for routine edits. Switch with /model, and adjust thinking with /effort.

Privacy and data

Claude Code runs on your machine, but the model does not. Your prompts and the file contents it reads are sent to Anthropic over TLS. What happens next depends on your account:

• Free, Pro, Max and Ultra: training on your data only if you allow it in privacy settings. Retention is five years if allowed and 30 days if not.
• Team, Enterprise, API and third-party platforms: Anthropic says it does not train on code or prompts sent to Claude Code under commercial terms, unless the customer opts in. Standard retention is 30 days.

Session transcripts are stored in plain text under ~/.claude/projects/ on your machine for 30 days by default. If a repo holds secrets, add deny rules for files like .env, and see the best-practices page for the exact syntax. Do not point Claude Code at client code until you have read your contract and the data-usage page.

Checklist

• Installed with the native installer, and claude --version prints a version
• Signed in with a plan or key you chose on purpose
• Working tree was clean before the first prompt
• Asked two or three read-only questions first
• Made one small change and read the diff before approving
• Named a test, lint or build command for Claude to run
• Know Esc, /rewind, /clear and Shift+Tab
• Ran /init and trimmed CLAUDE.md to what Claude cannot see in the code
• Checked your data-use setting for your account type
• Looked at /usage after the first long session

Next steps

You now know the loop: ask, read, change, check, commit. To practice it on a real build, follow the Claude Code tutorial. To make it behave across a team, read Claude Code best practices. To wire it to outside tools, see how to install an MCP server in Claude Code. If you are still choosing an agent, start at AI coding agents.

What this guide uses