Claude Code Commands Cheat Sheet — Full Reference

This article covers every Claude Code command, shortcut, and CLI flag — organized by how often you actually use them. Everything here is sourced from official Anthropic documentation and verified against v2.1.72 (March 2026).

Top 10 Commands You'll Use Every Day

These are the commands I reach for daily. Learn these ten and you can handle most workflows.

1. /compact — Compress Context

/compact
/compact Focus on TypeScript type definitions

Compresses the conversation when the context window gets long. You can optionally tell it what to focus on. Essential for long sessions and directly reduces token consumption.

When to use it? After 30+ minutes in a session, or as soon as you see the "context is getting large" warning. During a refactor, try /compact Focus on the auth module and current test failures to keep the most relevant context after compression.

2. /model — Switch Models

/model sonnet
/model opus

Switch between models based on the task — Sonnet for research, Opus for writing and complex reasoning. In the selection screen, use left/right arrow keys to adjust the effort level (thinking depth). Changes take effect immediately without interrupting the current response.

In practice: Use Sonnet for file exploration and grep searches, then switch to Opus for architectural decisions and code implementation. The Alt+P shortcut lets you switch models without clearing your current prompt — useful when you start typing and realize you need a more capable model.

3. /cost — Check Token Usage

/cost
/stats

/cost shows API token consumption for the current session. However, Pro/Max subscribers won't see /cost — it's designed for API users. If you're on a Pro plan, use /stats instead to view daily usage, session history, and streaks.

Build the habit: Check your token usage before ending every session. API users use /cost, Pro users use /stats. Once you know "this task used X resources," you can make better decisions next time — like "this part could have been done with Sonnet instead of Opus."

4. /diff — Review Changes

/diff

Opens an interactive diff viewer. Use left/right arrows to switch between git diff and individual Claude turn diffs, and up/down to browse files. Essential before committing.

Best practice: After asking Claude to refactor a module, run /diff before committing. Switch to "per-turn diffs" to see what changed at each step — this makes it easy to spot unintended modifications that can slip through in large refactors.

5. /clear — Reset Context

/clear

Clears conversation history and frees context. Also available as /reset or /new. Use it at task boundaries to save tokens for the next task.

/compact vs /clear: /compact summarizes and keeps context (maintains awareness). /clear is a hard reset (fresh start). Use /clear when switching to a completely different task in the same project. Use /compact when continuing the same task but the context is getting heavy.

6. Shift+Tab — Toggle Permission Modes

Press Shift+Tab to cycle through Auto-Accept Mode, Plan Mode, and normal mode.

• Normal mode: Asks for confirmation before every tool execution
• Auto-Accept: Runs all tools without confirmation (for trusted tasks)
• Plan Mode: Read-only access. Code changes are proposed as plans and require approval before execution

Alt+M does the same thing.

Practical workflow: Keep Auto-Accept on for writing tests or generating boilerplate. Switch to Plan Mode when touching production-critical files (deployment configs, package.json, DB migrations). The "show plan first → approve → execute" flow prevents accidents.

7. ! — Bash Mode

! npm test
! git status
! ls -la

Prefix input with ! to run bash commands directly. Executes without Claude's interpretation, and the output is added to conversation context. Press Tab to autocomplete from previous ! commands.

Why it's useful: Running git status through Claude means it interprets → executes → explains — three steps. ! git status runs directly and the output just enters the context. Use it when you want Claude to see the output but don't need it to interpret the command itself.

8. /resume — Resume Sessions

/resume
/resume auth-refactor

Resume a past session by ID or name. Without arguments, opens a session picker. Alias: /continue. From the CLI, use claude -c (most recent) or claude -r "name".

Example workflow: Pause feature development to handle a hotfix. After the fix, /resume to open the picker and jump back to the dev session. Name your sessions with /rename auth-refactor so they're easy to find later.

9. /memory — Manage CLAUDE.md

/memory

Edit CLAUDE.md memory files, enable or disable auto-memory. The core feature for maintaining project rules and context across sessions.

What to put in it? Rules like "use Vitest for tests," "CSS with Tailwind v4," "validate API responses with Zod." Any project-specific tech stack choices or coding conventions that you'd otherwise repeat every session.

10. @ — File Path Completion

Type @ during input to trigger file path autocomplete. Useful when referencing specific files in your questions.

Example: @src/components/Header.tsx Fix the responsive layout — Claude knows exactly which file to look at without searching. Saves tokens and gets faster results.

Session Management

Commands for starting, ending, resuming, and branching sessions. These become essential when you're juggling multiple tasks in parallel.

Cost & Performance Management

Commands for token consumption and model configuration. Mastering these six commands is the single most important thing for keeping your Claude Code bill under control.

Code Operations & Review

Reviewing changes, rewinding, and security checking. Always review before committing — especially after asking Claude to make large-scale changes.

Project Settings

Project initialization, configuration, and permission management.

External Integrations

IDE, browser, GitHub, and remote session integrations. The /remote-control feature is particularly interesting — it lets you control your local Claude Code from claude.ai, meaning you can direct work from your phone while away from your desk.

Keyboard Shortcuts

Some keys differ by OS and terminal. Switch tabs to see your platform.

CLI Flags

Flags you can specify when launching Claude Code. Especially important for scripting and automation.

Common use cases:

# Get a repo overview in one command and save to file
claude -p "Explain this repository's structure" > repo-overview.txt

# Work in an isolated git worktree without affecting your main branch
claude -w experimental-feature

# Automate code review in CI/CD
claude -p --output-format json --max-turns 5 "Review this PR"

# Allow only read-only tools for safe automated analysis
claude -p --allowedTools "Read" "Grep" "Glob" "Find issues in this code"

Commonly Used Flags

Automation & SDK Flags

Security & Permission Flags

Prompt & Configuration Flags

Other Flags

Full Command Reference

Alphabetical list of every slash command available in Claude Code. Some commands may not be visible depending on your plan, platform, or environment.

MCP servers can expose prompts as commands in the format /mcp__<server>__<prompt>, dynamically discovered from connected servers.

Bundled skills like /simplify, /batch, and /debug also appear alongside built-in commands when you type /. You can create your own skills as well.

Wrapping Up

Claude Code has over 50 commands, but you'll use about 10 of them daily. Start with /compact, /model, and /cost to build a habit of being aware of tokens and costs.

As you get comfortable, expand to /diff for pre-commit reviews, /fork for experimental branches, and /btw for quick side questions without breaking your flow.

For CLI flags, knowing --print and --output-format json opens up shell scripting and CI/CD integration possibilities.

Official documentation: CLI Reference | Interactive Mode