10 Minutes to Claude Code

中文版本

Claude Code is an AI coding agent.

It can read your project, edit files, run commands, and work through terminal, VS Code, JetBrains, desktop, web, and CI/CD workflows.

But you do not need to learn every surface on day one.

Start with the terminal.

This guide gets you through the first real use in about ten minutes.

1. Install

For macOS, Linux, and WSL, use the official install script:

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

On macOS, Homebrew also works:

brew install --cask claude-code

Windows PowerShell:

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

Windows CMD:

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

Windows also supports WinGet:

winget install Anthropic.ClaudeCode

After installation, enter a project:

cd ~/projects/my-app
claude

The first run prompts you to log in through your browser.

2. Do Not Edit Immediately

Start by asking it to read the project:

Read this project first. Do not edit files. Tell me what the project does, where the entry point is, how to start it, and how to test it.

This is a good first round.

You can check whether Claude Code understands the project without creating changes.

Then ask:

Which 5 files should a new contributor read first? Explain each one in one sentence.

Very often, the first value of Claude Code is not writing code. It is making an unfamiliar codebase readable.

3. Make One Small Change

After it reads the project, give it a small task:

Change the homepage button label from "Start" to "Try it now". Only change this text. Do not change styles. Tell me which file changed.

Or:

Add two tests for formatDate in src/lib/date.ts. Cover normal date and empty input. Do not change the implementation.

Do not begin with a large refactor.

Claude Code can handle complex work, but you should first learn how it plans, asks for permission, shows diffs, and runs verification.

4. Watch the Actions

Claude Code asks for confirmation before editing files.

You can approve individual changes, or switch to a more permissive mode for a session.

At the beginning, stay conservative.

Useful commands and actions:

| Goal | Command or Action |

| --- | --- |

| Show help | /help |

| Clear current context | /clear |

| Continue most recent conversation | claude -c |

| Resume an old conversation | claude -r |

| Ask once and exit | claude -p "explain this function" |

| Exit | exit or Ctrl+D |

| See commands and skills | Type / |

| Cycle permission modes | Shift+Tab |

For a one-off question:

claude -p "Explain what src/app/page.tsx does"

For a small project task:

claude "Fix the current lint error, then run lint to verify"

5. Add CLAUDE.md

Claude Code reads CLAUDE.md files in your project.

Think of it as a small instruction file for the agent.

Start with something short:

# Project Instructions

- Package manager: pnpm
- Before committing, run `pnpm lint` and `pnpm build`
- Keep changes scoped to the user request
- Do not rewrite unrelated files
- Prefer small, readable changes over broad refactors

This saves you from repeating the same constraints every session.

If you do not know where to start, ask Claude Code:

Generate a simple CLAUDE.md for this project. Only include real, verifiable project rules.

Then read it yourself before keeping it.

6. A Good First Workflow

Run the first session like this:

Read this project. Do not edit files. Summarize the structure, dev command, and test command.

Then:

Find one small improvement. Give me the suggestion first. Do not edit yet.

Then:

Implement only the first suggestion. Keep the edit scoped to related files. Run available verification. Do not commit.

Then:

Summarize what changed, which files changed, what verification ran, and what risk remains.

This may feel slower than asking for everything at once.

In practice, it usually saves time because you do not need to untangle a huge surprise diff afterward.

7. When Claude Code Fits

| Task | Fit |

| --- | --- |

| Reading an unfamiliar project | Great |

| Fixing a clear bug | Great |

| Writing tests | Great |

| Updating README or docs | Great |

| Small refactors | Good |

| Large technical migration | Be careful |

| Unclear product direction | Think first |

Claude Code is best for tasks where the goal is clear but the execution is tedious.

Examples: fixing lint, adding tests, finding entry points, explaining errors, updating docs, or cleaning up a small piece of old code.

If you do not know what you want yet, it will still try hard. The result just may not be the right thing.

8. My Suggestion

Do not treat Claude Code like a wish machine.

Treat it like a fast teammate who reads a lot, acts quickly, and needs clear boundaries.

You own the direction.

It accelerates the work.

That is the comfortable way to use it.

Reference

1. Claude Code Overview
2. Claude Code Quickstart
3. Claude Code CLI Reference
4. Claude Code Common Workflows
5. Claude Code Best Practices