GuideClaude CodeBeginner

Install Claude Code and use it safely on your first project

Install Anthropic's terminal coding agent, sign in, point it at a project, and learn to keep its edits under your control before you hand it real work.

Curated by Javier Díaz · Written with Claude from Anthropic’s documentation · Checked 18 October 2026 · 25 min

Claude Code is Anthropic's coding agent. Rather than answering questions in a chat window and leaving you to copy the results across, it works inside your project: it reads files, runs commands, edits code and can handle Git for you. The main version runs in your terminal, and the same engine sits behind the VS Code and JetBrains extensions, a desktop app and a browser version.

This guide is for developers and technical people who have used a terminal before and want a careful first session. It covers installation, signing in, getting Claude to explain a codebase, making and reviewing a small change, choosing how much freedom it has, and writing a CLAUDE.md file so you stop repeating yourself.

It earns its keep on work that spans several files or needs commands run and checked: tracing a bug, adding tests, tidying a module, writing up how something works. For a one-line question about syntax, an ordinary chat assistant is still quicker.

1. 01Install Claude Code

   Anthropic recommends its native installer, which keeps itself up to date in the background. Run the line for your system.

   macOS, Linux or WSL:

   curl -fsSL https://claude.ai/install.sh | bash
   

   Windows PowerShell (your prompt starts with PS):

   irm https://claude.ai/install.ps1 | iex
   

   If you prefer a package manager, brew install --cask claude-code and winget install Anthropic.ClaudeCode also work, but those copies do not update themselves, so you will need to upgrade them by hand.

   The installer prints nothing while it downloads, so give it a minute. Then open a new terminal window and check it worked:

   claude --version
   

   If the shell says claude is not recognised, the install folder is not on your PATH yet; Anthropic's troubleshooting page covers the fix. claude doctor gives a fuller health check without starting a session.

2. 02Start it in your project and sign in

   Claude Code works on whichever folder you launch it from, so move into the project first:

   cd path/to/your-project
   claude
   

   On the first run it opens your browser to sign in. Use the same account as your Claude subscription, or your Console account if you are paying per use. Credentials are stored, so you will not be asked again on that machine. Inside a session, /login switches accounts, and on a subscription /status helps you keep an eye on how much usage you have left.

   One trap: if your environment has an ANTHROPIC_API_KEY variable set and you approve it when asked, Claude Code uses that key instead of your subscription, and the usage is billed to the API account. On Pro and Max, usage limits are shared between Claude Code and the regular Claude apps.

3. 03Ask it to explain the code before changing anything

   Start with questions. Claude reads whatever files it needs, so you do not have to paste code in. Type plain requests at the prompt, for example:

   • give me a two-paragraph summary of what this repository does and how it is laid out
   • walk me through what happens when a user submits the signup form, file by file
   • which parts of this code have no tests?

   Type @ followed by part of a filename to point Claude at a specific file. This is also a cheap way to judge its accuracy on code you already understand before you trust it with code you don't.

4. 04Choose a permission mode

   The permission mode decides what Claude may do without asking you. The ones that matter at first:

   • Manual (config value default): Claude can read freely but asks before editing files or running most commands. The right choice while you are learning.
   • Accept edits (acceptEdits): file edits and basic file commands such as mkdir and mv inside the project go ahead; other commands still ask. Useful once you are happy reviewing changes afterwards with git diff.
   • Plan (plan): Claude investigates and writes a plan but cannot edit your code until you approve the plan.
   • Auto (auto): a separate safety model reviews actions instead of you and blocks ones that look risky. Anthropic is clear that this reduces prompts rather than guaranteeing safety.

   Recent versions start terminal sessions in auto mode by default, so check the status bar under the prompt. Press Shift+Tab to cycle modes; from auto, the first press takes you to Manual. You can also choose at launch:

   claude --permission-mode manual
   

   There is also a mode that skips all checks; leave it alone unless you are working inside a throwaway container or VM.

5. 05Make a small change and approve it

   Give Claude a contained task with an obvious right answer, such as the date in the footer is hard-coded; make it use the current year.

   In Manual mode it shows you each proposed edit or command and waits. You can choose:

   • Yes to allow that one action.
   • Yes, and don't ask again to allow similar actions. For file edits this lasts until the session ends; for a shell command it is saved as a rule for that repository.
   • No to refuse. Press Tab on an option first if you want to add a note, such as why you said no or what to do instead.

   When it finishes, look at the result as you would a colleague's change: run git diff, run the tests, open the page. You can also ask Claude to run the tests itself and show you the output.

6. 06Interrupt, undo and redirect

   You do not have to wait for Claude to finish a wrong turn.

   • Press Esc to stop it mid-task. Work done so far is kept, and you can type a correction.
   • Press Esc twice with an empty prompt, or run /rewind, to go back to an earlier point. You can restore the code, the conversation or both.
   • Run /clear before starting an unrelated task, so old context does not muddy the new one.

   Rewind only covers edits Claude made with its own file-editing tools. Files changed by shell commands, by you, or by another session are not tracked, so keep committing to Git and treat that as your real safety net.

7. 07Give it standing instructions with CLAUDE.md

   Claude starts each session with no memory of the last one, apart from a file called CLAUDE.md in your project root, which it reads every time. Use it for things you would otherwise keep repeating: how to build and test, conventions, where things live, what not to touch.

   Run /init and Claude will draft one from your codebase. Then edit it down. Short, checkable lines work best, for instance:

   # Commands
   - Run tests with `npm test`; run a single file with `npm test -- path/to/file`
   
   # Conventions
   - TypeScript strict mode; no `any`
   - Database migrations are hand-written; never edit files in `migrations/`
   

   Commit it so your team shares it. Personal preferences go in CLAUDE.local.md (add it to .gitignore) or ~/.claude/CLAUDE.md for all your projects. Anthropic suggests keeping each file under about 200 lines, as long files are followed less reliably. If your repository already has an AGENTS.md and no CLAUDE.md, Claude Code reads that instead. Run /context to confirm what was loaded.

8. 08Pick a sensible first real task

   Good first jobs are useful, limited in scope, and easy to check:

   • Write tests for one untested function, run them and fix any failures.
   • Explain and fix a specific error message you can paste in.
   • Update the README's setup section to match what the project actually needs.
   • Rename something confusing across the codebase and run the build to prove nothing broke.
   • Draft a commit message for your current changes, or ask what you have changed since the last commit.

   For anything touching several files, switch to plan mode first (Shift+Tab, or start a single prompt with /plan), read the plan, correct it, and only then let Claude implement. Avoid starting with vague requests like "improve the code" or with anything involving production credentials.

Written by Claude from the official pages above and checked against them on 18 October 2026. Products change; if a step no longer matches what you see, the sources have the current version. How we work