Claude Code Command Cheatsheet

⚙️

Installation & Setup

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

Install Claude Code globally via npm

claude

Start Claude Code (will prompt for authentication on first run)

claude update

Update Claude Code to the latest version

claude --version

Check your Claude Code version (confirm you're on latest)

claude config

Configure Claude Code settings

🚀

Basic Usage

claude

Start interactive REPL session

claude "explain this project"

Start REPL with initial prompt

claude -p "query"

Run one-off query, then exit (print mode)

cat file.py | claude -p "explain this code"

Process piped content

claude --continue

Continue most recent conversation

claude --resume

Show conversation picker to resume a session

/init

Initialize project with CLAUDE.md guide

/clear

Clear conversation history

/compact

Compact conversation to keep context window lean

/compact Focus on the authentication system

Compact conversation with custom focus instructions

/cost

Show token usage statistics

/memory

Edit CLAUDE.md memory files

/prime

Load key files and information into context window

! command

Execute bash commands directly

@ file/path

Reference specific file paths

# memorize this information

Add information to memory files

Escape (double-tap)

Undo/edit previous prompt

Shift + Tab

Toggle auto-accept mode for edits

Ctrl + R

Enable verbose output mode

\ + Enter

Create newline in input

🎛️

CLI Flags

claude -p "query" --output-format json

Output response as JSON (useful for scripting)

claude -p "query" --output-format stream-json

Stream JSON output in real-time

claude --verbose

Enable verbose logging

claude -p --max-turns 3 "query"

Limit number of agentic turns

claude --mcp-debug

Debug MCP server configuration issues

claude --dangerously-skip-permissions

Skip all permission prompts (use with caution!)

⚡

Slash Commands (Interactive Mode)

/config

View/modify configuration

/review

Request code review

/pr_comments

View pull request comments

/doctor

Check Claude Code installation health

/login

Switch Anthropic accounts

/logout

Sign out from your Anthropic account

/vim

Enter vim mode for alternating insert and command modes

/terminal-setup

Install Shift+Enter key binding for newlines (iTerm2 and VSCode only)

/bug

Report bugs (sends conversation to Anthropic)

/help

Get usage help

🛠️

Custom Slash Commands

echo "Analyze performance and suggest optimizations:" > .claude/commands/optimize.md

Create project-level custom command (available as /project:optimize)

echo "Review code for security vulnerabilities:" > ~/.claude/commands/security-review.md

Create global custom command (available in all projects)

echo "Fix issue #$ARGUMENTS with tests and PR description" > .claude/commands/fix-issue.md

Create parameterized command (use with /project:fix-issue 123)

# Add instant memory

Start input with # to quickly add to memory files

🎯

Advanced Workflows

claude -p "Find all TODO comments and fix them" --max-turns 10

Automated task with multiple turns

for file in *.py; do claude -p "Add type hints:" < "$file"; done

Process multiple files with Claude

git worktree add ../project-feature-a -b feature-a

Create Git worktree for parallel Claude Code sessions

claude -c -p "Continue debugging from yesterday"

Continue previous session in print mode

claude -p "think deeply about implementing OAuth2" --output-format json

Extended thinking with JSON output for parsing

🔌

MCP & Integrations

claude mcp

Configure Model Context Protocol servers

claude --mcp-debug

Launch with MCP debugging enabled

MCP Integration:

Add servers to .mcp.json file for team sharing
Use project config for repository-specific tools
Claude Code functions as both MCP server and client
Examples: Puppeteer, Sentry, GitHub, filesystem tools

💬

Prompt Engineering Tips

be concise

Force short, focused answers from Claude

read the entire file first

Force Claude to read complete files before responding

think deeply about this

Trigger extended reasoning and deeper analysis

think harder

Intensify Claude's reasoning process

make a plan first, don't code yet

Force planning phase before implementation

break this into small testable steps

Get incremental, verifiable progress

🎯

Complete Plan Mode Guide

What is Plan Mode?

A special operating mode where Claude Code only reads, analyzes, and plans—no file edits, commands, or commits
Mirrors senior engineer workflow: understand first, plan second, build third
Used to load context window with all relevant information before any code changes

Shift + Tab (twice)

Enter Plan Mode - Claude restricted from making file changes or running commands

/prime

Load key files and information into context window before planning

Correct me if I'm wrong, but...

Validate assumptions and get neutral, unbiased feedback

write a detailed specification for this feature in specs/

Create comprehensive plan/spec file for complex features

analyze the codebase and create an implementation plan

Have Claude understand and plan before building

Core Principles:

No Mutations: Cannot create, modify, or delete files (including plan.md/spec files)
Context Building: Read files and understand codebase, fill context window with key info
Ask Questions: Freely ask clarifying or "stupid" questions to validate assumptions
Validate Understanding: Use "Correct me if I'm wrong" for unbiased feedback

Recommended Workflows:

Plan → Execute (Quick Tasks): For small, simple changes
Plan → Write Spec → Execute (Mid/Large Features): Recommended for most work
In Plan Mode: analyze, create detailed specification
Review and iterate on the plan
Exit Plan Mode and execute according to the spec

Best Practices:

Be Explicit: Clearly state what you want planned - more explicit = better plan
Prime the Context: Use /prime to load key files before planning
Iterate on Plans: Ask follow-up questions to clarify and capture all requirements
Store Plans: Save specs in dedicated directory (e.g., specs/) for review and version control
Higher-Order Prompts: Treat plans as prompts for agents - pass specs as parameters
Parallelize Work: Use agentic loops for multi-agent, parallel development

Why Use Plan Mode?

Safety: No risk of accidental file changes or destructive actions
Clarity: Forces full understanding of problem and requirements before building
Efficiency: Reduces wasted time by ensuring right approach before execution
Scalability: Enables complex, multi-agent workflows with clear, shared plan

Plan Mode Summary Table

Step	Action/Advice
Enter Plan Mode	Shift+Tab twice
No Mutations	Only reads/analyzes, never edits files
Build Context	Load key files/info before planning
Ask Questions	Validate assumptions, use "Correct me if I'm wrong"
Plan → Execute	For quick tasks: plan, then build
Plan → Spec → Execute	For big tasks: plan, write spec, review, then build
Store Plans	Save specs in dedicated directory
Parallel Agents	Use agentic loops for multi-agent, parallel work
Higher-Order Prompts	Treat plans as prompts for agents
Review & Iterate	Always review and refine plans before execution

Key Takeaways:

The plan is the prompt: Well-crafted plan/spec is foundation for effective agentic coding
Think, then build: Always plan before coding - new standard for AI-augmented engineering
Set up codebase for agents: Organize files, specs, and context to maximize agent productivity

🤔

Plan Mode vs Auto Accept Mode

Shift + Tab

Toggle between Plan Mode and Auto Accept Mode

Plan Mode:

Claude asks for permission before each action
Best for: thinking, clarifying, and iterating before executing
Use when: working on critical code or unfamiliar tasks
Allows you to guide and course-correct Claude's approach

Auto Accept Mode:

Claude executes changes automatically without asking
Best for: trusted operations and repetitive tasks
Use when: you trust Claude's judgment on the task
Monitor the context usage indicator during auto mode

📋

Project Management & Planning

echo "# Attack Plan\n## Goals\n## Steps\n## Progress" > plan.md

Create markdown file for tracking plans and progress

use the plan.md file as a checklist and update progress

Have Claude maintain project progress in markdown files

consult multiple models for this critical decision

Get second opinions for important architectural choices

Project Management Best Practices:

Use markdown files (.md) for attack plans and progress tracking
Break big refactors into small, testable steps
Have Claude update progress in your planning documents
For critical decisions, consult multiple AI models (Claude, Gemini, GPT-4)
Keep a running checklist of completed vs pending tasks

⌨️

Keyboard Shortcuts & Navigation

\ + Enter

Quick escape to create newline

Option + Enter

Add newline (Meta+Enter with proper terminal config)

Shift + Tab

Toggle auto-accept mode

Escape

Interrupt Claude during any phase (thinking, tool calls, file edits)

Escape (double-tap)

Jump back in history and edit previous prompt

Arrow Keys

Navigate conversation picker when resuming sessions

Hidden Features:

View previous chats: Access via Claude Code UI in your current folder
Context usage indicator: Monitor context window fill in the UI
Double-tap Escape: Powerful feature for iterating on prompts
Use /compact before context window fills for best performance

🗑️

Dangerous Operations

⚠️ DANGER ZONE

The following commands can cause data loss or unexpected behavior. Use with extreme caution!

claude --dangerously-skip-permissions

🔥 Skip ALL permission prompts (Claude can modify/delete files without asking)

claude -p "Delete all test files" --dangerously-skip-permissions

🔥 Automated destructive operations without confirmation

rm -rf .claude/

🔥 Remove all Claude Code configuration and memory (PERMANENT!)

⚙️

Configuration Examples

Example settings.json configuration:

{ "permissions": { "allow": [ "Bash(npm run lint)", "Bash(npm run test:*)", "Read(~/.zshrc)" ], "deny": [ "Bash(curl:*)" ] }, "env": { "CLAUDE_CODE_ENABLE_TELEMETRY": "1", "NODE_ENV": "development" }, "theme": "dark", "auto_accept": false, "max_turns": 10 }

Configuration Hierarchy:

~/.claude/settings.json - Global user settings
.claude/settings.json - Project settings (checked into git)
.claude/settings.local.json - Local overrides (not checked in)
Enterprise policies take precedence over all user settings

💡

Best Practices

Effective Usage Tips:

Use /init to create CLAUDE.md for better context
Use /compact frequently before context window fills
Master Plan Mode: Enter with Shift+Tab twice for safe analysis and planning
Ask Claude to make a plan before coding with "make a plan first, don't code yet"
Use auto-accept mode (Shift+Tab) for trusted operations
Create custom slash commands for repeated workflows
Use Git worktrees for parallel Claude sessions
Monitor costs with /cost command and context usage in UI
Start small and scale up for complex tasks
Use think deeply for complex architectural decisions
Use MCP servers for extended functionality
Use markdown files for project planning and progress tracking
Break big refactors into small, testable steps
Use "be concise" for focused answers, "read the entire file first" for complete analysis
Double-tap Escape to edit and iterate on previous prompts
Consult multiple AI models for critical architectural decisions
Follow Plan → Spec → Execute workflow for complex features
Store specifications in dedicated specs/ directory
Use "Correct me if I'm wrong" to validate assumptions in Plan Mode