🧠 How to Use Claude Commands (With Examples That Actually Help)

TL;DR — Too Long Didn’t Read

• Claude.ai: use / menu, Projects, and Extended Thinking

• Claude Code has 15+ built-in commands for session, model, and code control

• /plan before big tasks, /diff before accepting changes, /compact when sessions get long

• Custom commands turn any repeated prompt into a one-word shortcut

Two Claude products, Two Command Systems

Most people use Claude the same way every day. Open chat, type a prompt, wait. That's fine. But Claude has a full command system that most users have never touched. These commands save time, cut repetition, and let you control exactly how Claude works. This guide covers the ones actually worth knowing, split by where you use them.

Claude has two main interfaces and they work very differently.

• Claude.ai is the chat interface you use in a browser or the desktop app. It has a basic / menu for triggering features like web search or file upload. Not much else.

• Claude Code is a separate terminal-based tool for coding and agentic tasks. It has a full slash command system, 20+ built-in commands, and lets you create your own. This is where the real power is.

You need a Claude Pro plan ($20/month) for full Claude.ai access. Claude Code requires a separate install (npm install -g @anthropic-ai/claude-code) and a Max or API account.

Before You Start: Get Set Up in 5 Minutes

If you use Claude.ai (browser or desktop app):

1. Make sure you're on a paid plan (Pro at $20/month or higher)

2. Open any chat and type / in the message box to confirm the menu appears

3. That's it. All Claude.ai commands in this guide work immediately

If you want to use Claude Code (terminal):

1. Make sure you have Node.js installed on your computer

2. Run npm install -g @anthropic-ai/claude-code in your terminal

3. Run claude to start a session

4. Type /help to see all available commands

5. Note: Claude Code requires a Max plan ($100/month) or API access

If you're not a developer and the terminal feels unfamiliar, skip the Claude Code section for now and start with Claude.ai. You can always come back.

Commands for Claude.ai (Chat)

These aren't slash commands, but they're features most users miss entirely.

1. Type / in the chat box

Opens a quick menu to trigger web search, upload files, or start a deep research task without digging through settings.

2. Use Projects to set persistent context

Not a command but works like one. Create a Project, add a system prompt, and every chat inside it inherits your instructions. Stop re-explaining your role, tone, and format every session.

3. Extended Thinking

Click the model selector and enable Extended Thinking on Opus 4.6. Claude reasons through the problem before answering. Use it for anything complex: analysis, strategy, debugging logic errors.

4. / + Style selector

In the chat input, you can switch Claude's writing style mid-conversation. Useful when you want a draft in one tone and a summary in another.

Commands for Claude Code (Terminal)

These are true slash commands. Type them directly in your Claude Code session.

Session management

5. /clear

Wipes conversation history and starts fresh. Use this every time you switch tasks. Old context bleeds into new responses and quietly degrades output quality.

6. /compact

Compresses long conversation history without losing important context. Use when your session gets long but you're not done yet. You can focus it: /compact keep the error handling patterns.

7. /context

Shows a visual breakdown of how much of your context window is used. Run this before things get slow. If you're over 80%, run /compact.

8. /cost

Shows token usage and estimated cost for the current session. Essential if you're on a usage-based API plan and want to avoid surprises.

Workflow and model control

9. /model sonnet or /model opus

Switches the AI model mid-session without starting over. Use Sonnet for fast, routine tasks. Switch to Opus when you hit something complex. Switch back when you're done.

10. /fast

Toggles fast mode. Same model, optimized for speed. Good for quick iterations and live debugging when you don't need maximum quality.

11. /plan

Puts Claude into planning mode. It maps out what it's going to do before taking any action. Use before any task that touches multiple files or has real consequences.

12. /diff

Opens an interactive viewer showing every change Claude has made in the session. Run this before you accept anything. Much better than trying to track edits mentally across files.

13. /memory

Opens Claude's persistent memory for the project. Claude saves things like architectural decisions and your preferences to a MEMORY.md file. This command lets you view and edit it directly.

Review and code quality

14. /simplify

Runs a three-agent review pipeline on your recently changed files. Flags duplicate logic, architectural issues, and inefficiencies, then fixes them. Run before any PR.

15. /review [PR number]

Pulls a GitHub pull request and reviews it for bugs and security issues. Requires the GitHub CLI to be installed and authenticated. Fastest code review you'll ever do.

16. /debug

Activates a focused debugging workflow. Claude analyzes the current error, traces it through relevant files, and proposes a fix, rather than just guessing at the problem.

Build your own

17. Custom commands

Create a file at ~/.claude/commands/review.md, write a prompt inside it in plain English, and type /review in any session to trigger it. The filename is the command. Works across all your projects. No coding required.Step 4 — Schedule it to run on a set schedule

Claude Command Limitations

Claude.ai's command system is thin. The real value is in Claude Code, which requires terminal comfort and a separate install.

Custom commands only work in Claude Code, not in the browser chat. The /review PR command requires the GitHub CLI and won't work without it set up first.