Claude Code Cheat Sheet

import { Aside, Card, CardGrid } from ‘@astrojs/starlight/components’;

Quick Reference — All Commands

CLI Commands (Terminal)

Command	Purpose	Example
claude	Start interactive REPL	claude
claude -p "prompt"	One-shot mode	claude -p "explain this error"
claude --model sonnet	Start with specific model	claude --model opus
claude --version	Check installed version	claude --version
cat file | claude -p "..."	Pipe mode — feed file content	cat log.txt | claude -p "summarize"
claude config	Manage configuration	claude config
claude -c / claude --continue	Continue most recent conversation	claude -c
claude -r / claude --resume	Resume specific session by ID/name	claude -r "auth-refactor"
claude update	Update to latest version	claude update
claude mcp	Configure MCP servers	claude mcp add my-server
claude doctor	Diagnose installation issues	claude doctor
claude -p --output-format json	JSON output for automation	claude -p --output-format json "list files"
claude -p --max-turns 3	Limit agentic turns	claude -p --max-turns 3 "fix lint"
claude -p --max-budget-usd 5	Set dollar spending limit	claude -p --max-budget-usd 5 "refactor"
claude --allowedTools "Read" "Bash(git *)"	Restrict tool access	claude --allowedTools "Read"
claude --system-prompt "..."	Custom system prompt	claude --system-prompt "You are a reviewer"

Use **one-shot mode** (`-p`) for quick questions. Use **REPL mode** for multi-turn conversations and complex tasks.

Slash Commands (Inside REPL)

Command	What It Does	When to Use
/help	List all available commands	Forgot syntax? Start here
/init	Create CLAUDE.md for project	First time in a new project
/compact	Compress conversation history	Context at 50-70% — frees 30-70% tokens
/compact [focus]	Compact with focus instruction	/compact Preserve the auth decisions
/clear	Erase ALL context, start fresh	Switching to unrelated project
/cost	Show token usage and estimated cost	Every 20-30 minutes to track spending
/model	Switch AI model mid-session	Change between Haiku/Sonnet/Opus
/status	Show version, model, account info	Quick session overview
/context	Visual context usage grid	See how full your context window is
/memory	Edit CLAUDE.md memory files	Update project memory without leaving session
/resume	Resume session or show picker	Return to a previous conversation
/doctor	Check installation health	Diagnose issues with your setup
/terminal-setup	Install terminal keybindings	Enable Shift+Enter for multi-line input
/vim	Enable vim mode	For vim users
/bug	Report a bug	Send bug report to Anthropic
/export	Export conversation to file	Save session for later reference
/exit or Ctrl+C	Exit session	Done working

`/compact` preserves task context while freeing space. `/clear` erases everything. Prefer `/compact` during active work.

Keyboard Shortcuts

Shortcut	Action
Escape	Cancel current generation
Escape × 2	Rewind — show message selector to undo
Ctrl+C	Interrupt / exit
Ctrl+V	Paste image from clipboard (macOS: Ctrl+V, NOT Cmd+V)
Ctrl+G	Open prompt in external editor ($EDITOR)
Ctrl+L	Clear terminal screen
Ctrl+O	Toggle verbose output
Shift+Tab	Cycle permission modes (Normal → Auto-Accept → Plan)
Option+T / Alt+T	Toggle extended thinking
Option+P / Alt+P	Switch model
Tab	Accept prompt suggestion
Up / Down	Navigate prompt history

Run `/terminal-setup` first to enable `Shift+Enter` for multi-line input in supported terminals (iTerm2, Ghostty, Kitty, WezTerm).

Input Shortcuts

Method	How
Multi-line input	\ + Enter, or Shift+Enter (after /terminal-setup)
Bash mode	! npm test — run shell commands without AI interpretation
File reference	@src/utils/auth.js — autocomplete with Tab
Paste image	Ctrl+V (clipboard screenshot)

Mode Selection Guide

Scenario	Mode	Command
Multi-turn debugging	REPL	claude
Quick question	One-shot	claude -p "..."
Review a diff	Pipe	git diff | claude -p "review"
Process a log file	Pipe	cat errors.log | claude -p "analyze"
CI/CD integration	One-shot	Script with -p flag
Learning / exploring	REPL	claude

Model Selection

Model	Cost	Best For
Haiku	$	Formatting, typos, simple edits, quick answers
Sonnet	$$	Features, debugging, code review, documentation
Opus	$$$	Architecture, complex debugging, security, novel problems

Default to **Sonnet** for most tasks. Use **Haiku** for simple edits. Reserve **Opus** for complex architecture and reasoning.

---

Best Practices for CLAUDE.md

The 6 Essential Sections

Every project CLAUDE.md should include these sections:

# Project: [Name]

## 1. Project Overview
Tech stack: Node.js 20, Express, PostgreSQL, Redis
Architecture: Monorepo with packages/ directory

## 2. Architecture Rules
- Routes → Services → Repositories → Database
- All business logic in services, never in routes
- Shared types in packages/shared/

## 3. Coding Conventions
- Files: kebab-case (user-service.ts)
- Classes: PascalCase (UserService)
- Functions: camelCase (getUserById)
- Error handling: Promise<Result<T, AppError>>
- Imports: Absolute via @/ prefix

## 4. Commands
npm run dev          # Start dev server (port 3000)
npm test             # Run all tests
npm run lint         # ESLint + Prettier check
npm run db:migrate   # Run database migrations

## 5. Constraints (DO NOT)
- Do NOT use `any` type — always define proper types
- Do NOT install dependencies without asking first
- Do NOT put business logic in route handlers
- Do NOT commit .env files
- Do NOT modify database schema without migration

## 6. Context (Tribal Knowledge)
- Redis keys MUST include version prefix: v1:user:123
- JWT expires in 15 min, refresh token in 7 days
- Legacy /api/v1 routes maintained for mobile app compatibility
- Payment gateway sandbox mode uses TEST_ prefix keys

CLAUDE.md Size Guide

Size	Tokens	Quality
300-500 words	~400-650	Minimal — misses edge cases
500-800 words	~650-1000	Sweet spot for most projects
800-1200 words	~1000-1500	Comprehensive — large projects
1200+ words	1500+	Too large — diminishing returns

Keep CLAUDE.md under 1000 words. Every token spent on project memory is a token NOT available for your actual work.

File Hierarchy

Location	Scope	Priority
~/.claude/CLAUDE.md	Global (all projects)	Lowest
./CLAUDE.md	Project root	Main — use this
./src/CLAUDE.md	Directory-level	Highest (overrides root)

---

File Patterns Guide

.gitignore — Protect Secrets

# Secrets — NEVER commit
.env
.env.local
.env.*.local

# Dependencies
node_modules/
vendor/
.venv/

# Build output
dist/
build/
.next/

# OS files
.DS_Store
Thumbs.db

.env.example Pattern — Safe for Claude

# .env.example — committed to git, safe for Claude to read
DATABASE_URL=postgresql://user:password@localhost:5432/dbname
REDIS_URL=redis://:password@localhost:6379
API_KEY=your_api_key_here
STRIPE_KEY=sk_test_placeholder

# .env — real secrets, gitignored, NEVER read by Claude
# DATABASE_URL=postgresql://realuser:s3cr3t@prod.db:5432/prod

**Never** let Claude read your `.env` file directly. Always use `.env.example` with placeholder values. If secrets are accidentally exposed, rotate them immediately.

Permission Quick Reference

Command Category	Action
ls, pwd, cat, head, tail	Approve freely — read-only
grep, find	Approve freely — search only
git status, git log, git diff	Approve freely — read-only git
git commit -m "..."	Review message, then approve
npm install <package>	Verify package name first
File writes in src/	Verify path and content
rm, git push --force	Always deny unless intentional
curl, wget	Always deny — data exfiltration risk
Operations on ~/.ssh/, ~/.aws/	Always deny — sensitive directories

Secret Scanner Setup (Gitleaks)

# Install
brew install gitleaks

# Add pre-commit hook
cat > .git/hooks/pre-commit << 'HOOK'
#!/bin/bash
gitleaks protect --staged --verbose
if [ $? -ne 0 ]; then
    echo "SECRETS DETECTED! Commit blocked."
    exit 1
fi
echo "No secrets detected."
exit 0
HOOK
chmod +x .git/hooks/pre-commit

---

Context Management

Token Cost Estimates

Content Type	Tokens per 1000 chars
English prose	~250
Source code	~400
JSON data	~400
Small file (<5KB)	1,000-2,000 total
Medium file (5-20KB)	2,000-5,000 total
Large file (>50KB)	10,000+ total

Context Health Monitor

Level	Quality	Action
0-50%	Excellent	Continue working
50-70%	Good	Plan to /compact soon
70-85%	Degrading	Run /compact now
85-95%	Poor	Must compact or clear
95%+	Unreliable	/clear and restart

Power Combos

Combo	Workflow	When
/cost → /compact	Check → compress	Routine maintenance (every 30 min)
/clear → /init	Reset → configure	Starting new project
/compact → work → /cost	Compress → continue → verify	Long sessions

---

Workflow Quick Starts

New Project Setup

cd my-project
claude               # Start session
/init                # Create CLAUDE.md
# Edit CLAUDE.md with 6 sections above
/cost                # Check baseline usage

Long Session Protocol

# Every 20-30 minutes:
/cost                # Check context level

# At 50-70% context:
/compact             # Free space, keep task context

# Switching to unrelated task:
/clear               # Full reset

Safe Code Review

git diff main | claude -p "Review this diff for bugs, security issues, and style"

Context-Efficient Exploration

# Step 1: Overview first
"Show me the directory structure of src/"

# Step 2: Targeted reading
"Show me function signatures in src/auth/"

# Step 3: Deep dive only where needed
"Read src/auth/login.ts — I need to understand the token flow"

Enjoying this guide?

Get weekly AI coding tips and be the first to know about new modules.

Subscribe