Claude Code 101 Basics

10 examples to get you started with Claude Code 101 - 7 basic and 3 intermediate.

Prerequisites

• A terminal you are comfortable working in, and a project directory you want Claude Code to work on.
• The Claude Code CLI installed on your machine (covered in example 1 below).
• An Anthropic account with access to Claude Code, since the CLI needs to authenticate before it can run a session.

Basic Examples

1. Install the CLI

Install Claude Code from your terminal before doing anything else.

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

• The CLI installs globally so you can launch it from any project directory.
• Installation is a one-time step; after this you only need to keep it updated occasionally.
• If npm is not your preferred package manager, check platform.claude.com/docs for the current install options.

Related: Installing Claude Code and Connecting Your First Project - the full walkthrough this example summarizes.

2. Check the installed version

Confirm the CLI installed correctly and see which version you are running.

claude --version

• A version number confirms the install succeeded and the claude command is on your PATH.
• Run this again after an update to confirm the new version took effect.
• If the command is not found, your shell's PATH likely needs to include the global npm bin directory.

3. Launch the REPL in a project

Start an interactive Claude Code session inside the project you want help with.

cd ~/projects/my-app
claude

• Claude Code launches as a REPL: a standing session where you type natural-language requests.
• It orients itself in the current directory, so always cd into the project first.
• The session stays open across multiple requests until you exit it.

Related: Navigating the Claude Code REPL: Prompts and Slash Commands - what you can type once the session is open.

4. Give it your first natural-language request

Type a plain-English instruction at the prompt instead of writing code yourself.

> Add a README.md that explains what this project does

• Claude Code reads the project first, then plans and writes the file.
• You do not need special syntax; describe the outcome you want.
• Vague requests get vague results, so name specifics (files, functions, behavior) when you can.

5. Review a proposed change

See what Claude Code wants to change before it happens.

# Claude Code shows a diff-style preview of the edit
# you approve, edit the request, or reject it

• Reviewing the diff is how you catch a misunderstood request before it becomes a real edit.
• Whether you see this preview by default depends on your permission mode.
• Treat the first few sessions in a new project as a chance to build trust in its plans.

Related: Permission Modes Explained: Auto-Accept vs Ask-Every-Time - controls whether you see this preview for every action.

6. Use a built-in slash command

Trigger built-in behavior with a leading slash instead of a natural-language request.

> /init

• Slash commands are distinct from natural-language requests; the leading / signals a built-in action.
• /init is a common first command in a new project, since it lets Claude Code build up context about the codebase.
• Other slash commands exist for clearing or compacting the session's context.

7. Exit the session cleanly

End your Claude Code session when you are done.

> /exit

• Exiting cleanly closes the REPL session without leaving background work running.
• Your project files retain whatever changes were already approved during the session.
• You can always relaunch with claude from the same directory to pick up where you left off.

Intermediate Examples

8. Set your permission mode for a session

Choose whether Claude Code asks before each action or proceeds on its own.

claude --permission-mode ask-every-time

• ask-every-time prompts for approval before each file edit or shell command.
• Omitting the flag uses your configured default, which may be auto-accept.
• Start new or unfamiliar projects with ask-every-time until you trust the agent's plans in that codebase.

Related: Permission Modes Explained: Auto-Accept vs Ask-Every-Time - the full comparison of both modes.

9. Research safely with Plan Mode first

Let Claude Code investigate a risky or unfamiliar change before it touches any files.

> /plan
> Investigate why the checkout flow fails intermittently and propose a fix

• In Plan Mode, Claude Code can read files and reason about an approach, but it cannot make real edits.
• Review the proposed plan, then switch to execution mode to let it actually apply the fix.
• This two-step flow is the safest way to hand off a task you do not fully understand yet.

Related: Switching Between Plan Mode and Execution Mode - the full workflow for moving from research to real edits.

10. Chain a multi-step task across files

Give Claude Code one instruction that requires touching several files and running a command.

> Rename getUserData to fetchUserProfile everywhere it's used, then run the test suite

• A single instruction can trigger reading several files, editing each usage, and running a shell command to verify the result.
• This is the agent loop in practice: read, plan, act, verify, all from one natural-language request.
• Reviewing the summary Claude Code reports back confirms the test suite actually passed, not just that files changed.

Related: The Claude Code Mental Model: Agent, Not Autocomplete - why one instruction can safely trigger a multi-step plan like this.

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.