Claude Code Tutorial for Beginners: Your First Week

Get Claude Code running in whichever surface suits you — the terminal CLI, the desktop app for Mac or Windows, claude.ai/code in the browser, or the VS Code/JetBrains extension. Sign in with your Claude Pro or Max account, or an API key if your organisation manages access centrally.

Then, before asking it to change anything, use it to learn your own codebase:

• Ask: “Give me an overview of this repository’s architecture.”
• Ask: “How does authentication work in this project? Walk me through the flow with file references.”
• Ask: “What would you flag as technical debt here?”

This does two things. It shows you how the agent explores code — searching, reading, synthesising — and it builds calibrated trust: you can judge its answers against a codebase you know. Resist making edits today. Day 1 is reconnaissance.

Today you make every future session better. Create a CLAUDE.md file at your repository root — Claude Code reads it automatically at the start of every session. Include:

• Commands: how to build, run tests, lint, and start the dev server.
• Architecture notes: key directories, where business logic lives, patterns to follow.
• Conventions: naming, error handling, test style — the things reviewers usually catch.
• Pitfalls: the flaky test, the file nobody should touch, the migration quirk.

A useful shortcut: ask Claude Code itself to draft the file by analysing the repo, then correct it — the corrections are exactly the tribal knowledge worth writing down. Keep it under a page; it is onboarding notes, not documentation. Commit it so your whole team benefits. Then spend the rest of the day on small, low-stakes edits — a docstring, a rename — to learn the permission and diff-review flow.

Now do actual backlog work. Pick a genuine bug or small feature and practise the delegation loop:

• Specify the outcome, not the method: “Users can currently submit the form twice; prevent duplicate submissions and add a regression test” beats “fix the form bug”.
• Demand verification: tell it to run the test suite and iterate until green. This closed loop — edit, run, read failures, fix — is where agentic coding earns its keep.
• Review the diff like a colleague’s PR: read every change, ask “why did you change this file?”, and request revisions in plain language.

Also learn the escape hatches today: interrupt a session going the wrong direction, ask the agent to explain its plan before executing, and start a fresh session when context gets muddled rather than arguing with a confused one. Knowing when to stop and restate is a core skill, not an admission of failure.

Claude Code is fluent in git, and using that fluency changes how safely you can delegate:

• Branch by default. Start tasks with “create a branch for this” so agent work stays isolated and disposable — a bad session costs you a branch delete, nothing more.
• Let it write commits. Claude Code produces well-structured commit messages that describe intent. Review them like you review code.
• Prepare pull requests. Ask it to draft the PR description summarising what changed and why, with test evidence. Reviewers consistently notice the difference.
• Use it for git surgery. Conflict resolution, cherry-picks, bisecting a regression — tasks many engineers dread are well-suited to an agent that reads diffs tirelessly.

By the end of Day 4, aim to have shipped at least one real merged change where Claude Code did the bulk of the work and you did the review. That milestone converts sceptics faster than any demo.

Friday is for making the tool yours:

• Tune permissions. Pre-approve safe, frequent commands (tests, linters, formatters) so sessions flow without constant prompts; keep approvals on anything destructive.
• Add a hook. Hooks run deterministic scripts on events — for example, auto-running the formatter after every edit. Policy enforced by code, not by prompting.
• Try a skill. Skills package repeatable workflows — a release checklist, a review procedure — invocable on demand.
• Connect one MCP server. The Model Context Protocol links Claude Code to external systems: your issue tracker, a database, a browser. Start with one integration you would use daily.
• Meet subagents. For larger tasks, Claude Code can delegate to parallel subagents — a taste of orchestrating work rather than doing it.

You now have a personalised, verifiable, extensible setup — the foundation everything advanced builds on. Our Claude Code overview maps the full feature surface.