Your first project

You've installed Claude Code. Time to use it on something real.

Pick a project

Start with an existing project you know well. A side project, a work project you have permission to experiment on, or even a throwaway repo. Don't pick something brand new — Claude Code is most useful when it can read and understand existing code.

cd ~/Projects/my-real-project
claude

You'll land in a prompt. Claude has already started reading your project structure in the background.

The CLAUDE.md file

Before you give Claude a task, add a CLAUDE.md file at the root of your project. This is how you give Claude persistent context about your codebase — conventions, architecture, commands, preferences.

Here's a minimal example:

# My Project

A Next.js 15 app for tracking book reading.

## Stack
- Next.js App Router, TypeScript strict
- Drizzle ORM + Postgres
- Tailwind + shadcn/ui

## Commands
- `npm run dev` — start dev server
- `npm run typecheck` — type check
- `npm test` — run tests

## Conventions
- Server Components by default, "use client" only when needed
- All forms use Server Actions, not API routes
- Drizzle schemas in src/db/schema/

Claude reads this on every session. The more specific you are, the better the results. Think of CLAUDE.md as onboarding documentation for a new engineer — except the engineer is very fast and doesn't forget.

Your first task

Start small. Pick something you'd normally do manually but it's boring:

• "Add JSDoc comments to every exported function in src/lib/"
• "Find all places where we still use the old useState pattern and migrate them to useOptimistic"
• "The CONTRIBUTING.md is out of date. Re-read the current codebase and rewrite it."

Type your prompt. Hit Enter. Watch what Claude does.

Understanding what you see

As Claude works, you'll see:

• Tool calls — [Reading src/lib/utils.ts], [Editing src/lib/date.ts]
• Reasoning — short summaries of what Claude is doing and why
• Diffs — proposed file changes, which you can approve or reject
• Command output — when Claude runs a shell command, you see the output inline

You stay in control. By default, Claude asks before edits and commands that touch the filesystem. You can loosen or tighten this in the session.

Review and commit

When Claude is done, don't skip review. Read the diffs. Run your tests. Check the code does what you actually wanted.

Then commit yourself — Claude Code can commit too, but when you're starting out, do it by hand. You want to stay close to the changes.

git diff
git add .
git commit -m "Add JSDoc comments to utility exports"

What you've learned

You can now:

• Install Claude Code
• Set up a project with CLAUDE.md
• Run an agent task
• Review and commit the output

That's the foundation. In the next lessons (coming soon), we'll go deeper: Skills and MCPs, subagents for parallel work, and production workflows.