Getting Started with Claude Code: From Install to First Shipped Feature

Claude Code is Anthropic's agentic coding tool: Claude, running in your terminal (or IDE, or browser), with permission to read your files, run your commands, and edit your code. The pitch is simple — instead of copy-pasting snippets into a chat window, you give Claude the same access a new teammate would get, and it does the work where the work lives.

This guide takes you from zero to a productive setup: installation, your first real session, the configuration that actually matters, and the habits that separate people who get senior-engineer output from people who get autocomplete.

1. Install it

Claude Code runs on macOS, Linux, and Windows. Install via npm:

npm install -g @anthropic-ai/claude-code

cd your-project
claude

On first run you'll authenticate with your Anthropic account — Claude Code works with Claude Pro/Max subscriptions or with an API key. Once you're in, you get a chat prompt that lives in your project directory. That directory scoping matters: Claude can see and touch this project, not your whole machine.

2. Your first session: pick a real task

Skip the toy demos. The fastest way to understand Claude Code is to hand it something you actually need done, then watch how it works. Good first tasks:

• "Explain this codebase: what are the main modules and how do they fit together?"
• "Find where user authentication happens and walk me through the flow."
• "This test is failing — figure out why and fix it." (paste nothing; it will find it)
• "Add input validation to the signup endpoint, following the patterns this repo already uses."

Notice what happens: Claude searches the repo, reads only the files it needs, proposes changes, and asks permission before running anything with side effects. You approve or reject each action — or grant standing permissions as trust builds. That permission system is the core of the tool. You can be as paranoid or as hands-off as your project deserves.

3. Create a CLAUDE.md — the five minutes with the highest ROI

CLAUDE.md is a markdown file at your project root that Claude reads at the start of every session. It's where you write down what you'd tell a new hire on day one. Run /init and Claude will draft one from your codebase; then edit it to add what only you know:

# MyApp

Next.js 15 (App Router) + TypeScript + Postgres via Drizzle.

## Commands
- npm run dev — dev server (port 3000)
- npm run test — vitest; run before claiming anything works
- npm run db:migrate — apply migrations

## Conventions
- Server components by default; "use client" only when interactive
- All DB access goes through src/db/queries/ — never inline SQL in routes
- Error messages are user-facing: no stack traces in responses

## Gotchas
- The payments webhook (src/app/api/stripe/) is load-bearing and
  under-tested. Do not refactor it casually.

Every instruction in this file is an instruction you never type again. When Claude does something wrong, don't just correct it in chat — ask yourself if the correction belongs in CLAUDE.md. Teams that treat this file like real documentation compound their setup; teams that skip it re-explain their stack every session.

4. Learn the four commands that matter

• /init — generates a CLAUDE.md from your codebase. Run it once per project.
• /clear — wipes the conversation. Use it between unrelated tasks; a context window full of the last task's details makes the next task worse.
• Esc — interrupts Claude mid-action. You're never a passenger; steer the moment it heads somewhere wrong.
• /help — everything else, discoverable when you need it.

5. Extend it: skills and slash commands

This is where Claude Code stops being a chat tool and starts being a platform. Skills are folders of instructions (a SKILL.md plus optional scripts) that teach Claude a repeatable capability — "how we do code reviews", "how to audit a site's security", "how we write commit messages". Slash commands are reusable prompts you invoke by name: /review, /test, /security-review.

You can write your own — or install a library of them. Our ClaudeThings kits package 89 agents, 103 skills, and 181 slash commands built exactly this way (and the free SKILL.md validator lints the ones you write yourself).

6. The workflow habits that actually matter

Plan before code on anything non-trivial

For multi-file changes, ask Claude to propose a plan first and critique it before letting it edit. Claude Code has a dedicated plan mode for exactly this. A wrong plan costs one message to fix; a wrong implementation costs a review cycle.

Make it verify its own work

End feature requests with "run the tests and fix whatever breaks" or "start the dev server and confirm the page renders." Claude Code can execute the feedback loop itself — that's the difference between code that looks right and code that is right.

Commit early, keep diffs reviewable

Let Claude make git commits as it goes (it writes good commit messages). Small commits mean you can review its work like a colleague's PR instead of auditing a 40-file surprise.

Scope your asks

"Fix the validation bug in signup" beats "improve the auth code." Claude treats vague scope as permission to be ambitious — which is occasionally wonderful and usually a bigger diff than you wanted.

7. Where people get stuck

• Trusting it blindly, then swearing off it entirely. Both extremes lose. Treat Claude Code like a very fast mid-to-senior engineer: review the work, especially anything touching auth, money, or data deletion.
• One giant session for everything. Context bloat degrades quality. /clear between tasks; one task, one conversation.
• Never reading the diff. The habit that catches 95% of problems is the one you already have: read the diff before it merges.
• Under-investing in CLAUDE.md. If you've corrected the same thing twice, it belongs in the file.

FAQ

Keep reading