Claude Code for Beginners: Your First Hour, Step by Step

Claude Code is an assistant that works inside your project: it reads your files, runs commands and edits code when you ask it to, in plain language. Your first hour needs eight steps, and none of them is hard: install it with one command, sign in, open it in a repository, ask it a question, have it make one small change, read the diff, commit, and decide what to learn in hour two. The one thing to get right from the start is reviewing what it did, so this walkthrough spends more time on that than on anything else.

Before you start

• An account. Claude Code needs a Claude Pro, Max, Team or Enterprise plan, or a Claude Console account; the free plan does not include it.
• A terminal. On a Mac, Terminal; on Windows, Windows Terminal or PowerShell. If you would rather not use one at all, the Desktop app runs the same engine with buttons; see Desktop app vs CLI vs VS Code.
• A small project that is already in git, with nothing uncommitted. A clean starting point is what makes every later step easy to check and easy to undo.

Step 1: Install it (two minutes)

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

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

# Then, in a new terminal window
claude --version

Those are the commands in Anthropic’s quickstart↗ (opens in a new tab). If claude --version prints a version number, you are done. On Windows there is a choice between a native install and WSL, and a few errors worth knowing; they are all in Claude Code on Windows.

Step 2: Sign in (one minute)

Type claude and press Enter. The first time, it opens your browser to sign in; approve it and come back to the terminal. You stay signed in after that. If you ever need to switch account, type /login inside a session; /status shows which account and model you are using.

Step 3: Open it in a repository

cd path/to/your-project
claude --permission-mode default

Always start Claude Code from the project folder: that folder is what it can see and work on. The first time you open a folder it asks whether you trust it; say yes only for code you know.

The flag on the second line matters for your first hour. Recent versions start interactive sessions in auto mode, where, per the permission modes page↗ (opens in a new tab), a classifier reviews actions instead of you and Claude edits most files and runs most commands without asking. That is efficient once you trust it. For now, --permission-mode default starts in Manual mode, so every edit and command waits for your yes. You can switch mode at any time with Shift+Tab.

Step 4: Ask a first question

Start by asking, not changing. You do not need to point it at files; it reads what it needs.

• “What does this project do, and how is it organised?”
• “Where is the entry point, and what happens when the app starts?”
• “How do I run the tests?”

Check its answers against what you know. It is usually right, and when it is not, you have learned early how much to trust it on this codebase. That calibration is the most useful thing you get from the first hour.

Step 5: Make one small change

Choose something you could do yourself in five minutes, so you can judge the result: fix a typo in an error message, add a missing test for a small function, rename a confusing variable in one file. Be specific about what and where:

In src/utils/format.ts, formatPrice(5.5) returns "5.5". It should always show two decimal places, "5.50". Fix it and add tests for 5.5 and 5.

In Manual mode each edit appears with a prompt. Read it and answer yes, no, or tell it what to do instead. When it wants to run a command, such as the test suite, it asks again. Saying no is normal and costs nothing; it adjusts and tries another way.

Step 6: Review the diff

When it says it is done, do not take its word for it. Type /diff: according to the interactive mode reference↗ (opens in a new tab), it shows the changes in your working tree, Claude’s edits alongside anything else you have not committed. Because you started clean, everything there is from this session. Check three things:

1. It changed only what you asked for, in the files you expected.
2. The test it wrote would fail without the fix. Ask it: “Would this test have failed before your change?”
3. The tests pass. Run them yourself, or ask it to and read the output.

If you do not like the result, press Esc twice with an empty prompt, or type /rewind, to go back to an earlier point. Checkpoints↗ (opens in a new tab) undo Claude’s own file edits but not files changed by shell commands it ran, which is one more reason to start from a clean git tree: git restore is your final undo. Stopping and rewinding in more depth: how to stop a Claude Code task.

Step 7: Plan mode, in one paragraph

For anything bigger than your first change, press Shift+Tab until plan mode is on, or start the prompt with /plan. Claude then reads and explores but does not edit; it writes a plan and waits for you to approve it, change it or keep planning. It is the cheapest way to catch a wrong approach before any file changes. Everything else about it is in Claude Code plan mode.

Step 8: Commit

You can commit yourself as usual, or ask: “Commit my changes with a descriptive message.” Either way, read the message and check what went in with git show --stat. A commit that says what changed and why is also what you will want in a week when you wonder what the assistant did.

Hour two: what to learn next

• Give it standing instructions. Run /init and Claude writes a starting CLAUDE.md with your build commands, test commands and conventions, which it reads at the start of every session. Keep it short. How it layers with memory: Claude Code memory.
• Manage context. Every message, file and command output fills the same context window. /clear starts fresh between unrelated tasks; the rest is in context engineering for Claude Code.
• Try auto mode on a low-risk task, once you know what good output looks like on your codebase, and compare how much you had to correct.
• Decide where you will work: terminal, editor or app. Claude Code in VS Code covers the editor route.
• Anthropic’s own best practices↗ (opens in a new tab) page is the next read; it is short and specific.
• Keep the commands cheat sheet open for a week, then work through 25 Claude Code best practices.

Keep a list of what it did

By the end of the first week you will have had dozens of sessions, and the terminal remembers none of them in a way you can scan. The habit that pays off early is keeping the work list outside the chat: a board where each task says what was asked, what changed and how it was checked. fenbs is built for that, with Claude Code as a member of the board. Connect it with claude mcp add --transport http fenbs https://fenbs.ai/api/mcp and /mcp, then follow a task-tracking workflow for Claude Code for the four lines of CLAUDE.md that make it read and update the board on its own.