Track 4

Claude Code

An agent that lives in your codebase

Use Claude as a coding agent in your terminal, IDE, or the cloud. Master plan mode, slash commands, Skills, hooks, MCP servers, and subagents.

WhereYour terminal (the CLI), VS Code & JetBrains extensions, the desktop app, or the cloud at claude.ai/code. Free with a Claude plan or API account.
claude — terminal
> add input validation to the signup form
● Read components/SignupForm.tsx
● Edit components/SignupForm.tsx
+ if (!email.includes("@")) setError("Invalid email")
- // TODO: validate
● Bash npm test
✓ 14 tests passed
Done — validation added and tests are green.

The big picture

An agent that lives in your codebase

Claude Code reads your repository, edits files, runs commands, and verifies its own work. You describe the change; it plans, executes, and iterates until it's done — right where you already work.

Understands the repo

Searches and reads across files to grasp context before it changes anything.

Edits & runs

Makes multi-file edits, runs tests and builds, reads the output, and fixes what broke.

Plans before acting

Plan mode investigates read-only and proposes an approach for you to approve.

Handles git

Writes meaningful commits, manages branches, and can open pull requests.

Extensible

Skills, hooks, MCP servers, and subagents let you shape exactly how it works.

Everywhere you code

Terminal, IDE, desktop, and cloud — the same agent across surfaces.

Your roadmap

From first run to a customized agent

Install it, learn the loop, then layer on plan mode, Skills, hooks, MCP, and subagents.

Get it running

  1. 1

    Install the CLI

    Use the install command from the official docs (for example npm install -g @anthropic-ai/claude-code, or the native installer). Prefer the IDE? Install the Claude Code extension for VS Code or JetBrains.
  2. 2

    Open a project and launch

    From your repo, run claude to start an interactive session. On first run you'll sign in through the browser.
  3. 3

    Describe a task in plain English

    Try something small: “add input validation to the signup form and write a test for it.” Claude reads the relevant files, makes edits, and runs the test.
  4. 4

    Review every change

    Claude shows diffs as it goes. Read them. Approve, or steer with feedback — “use the existing validation helper instead.”
  5. 5

    Create a CLAUDE.md

    Run /init to generate a CLAUDE.md — project memory where you record build commands, conventions, and gotchas Claude should always follow.

CLAUDE.md is your highest-leverage file

A few lines about how your project builds, tests, and what patterns to follow makes every future task sharper. Treat it as living documentation for your agent.

Keep it handy

Quick reference

Extension points

MechanismLives inAdds
CLAUDE.mdRepo rootPersistent project memory & conventions.
.claude/skills/Repo or home dirReusable, specialized task know-how.
settings.json hooks.claude/Shell commands that fire on events.
MCP serversclaude mcp addExternal tools & data sources.
.claude/agents/Repo or home dirCustom subagents for focused work.

Plan first

Shift+Tab into plan mode for anything unfamiliar.

Learn the slashes

Type / to discover commands in context.

Mind permissions

Auto-approve only what you're comfortable with.

Hook your checks

Lint and test automatically on every change.

Add MCP tools

Give the agent access to your systems.

Use subagents

Parallelize search, review, and big sweeps.

Common questions

Good to know

Previous trackClaude CoworkNext track AI Automation