Launch: The Comprehensive Claude Code Cheat Sheet — 200+ Items Curated

One-line install via npm. Requires Node.js 18+.

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

Navigate to your project directory and launch Claude Code.

cd your-project && claude

Auto-generates a CLAUDE.md with your project's structure, stack, and conventions. Essential first step.

/init

Toggle between Plan and Execute mode. ALWAYS plan first, then execute. The #1 tip from every expert.

Shift+Tab → describe what you want → Shift+Tab to execute

/clear resets the entire conversation. /compact summarizes and compresses context. Use /compact when context gets heavy, /clear for fresh start.

/compact  # summarize context
/clear    # full reset

Claude's built-in memory consolidation. Runs between sessions like REM sleep — prunes stale notes, resolves contradictions, organizes by topic. Replaces manual MEMORY.md management.

/dream

Add 'ultrathink' or 'think harder' to your prompt for complex architecture decisions. Uses extended thinking with more tokens.

ultrathink: Design the database schema for a multi-tenant SaaS app with row-level security

Made a mistake? Rewind to any previous point in the conversation and fork from there. Like git for your chat.

/rewind  # go back to a previous state

Expert consensus: start with 10-20 lines. Project name, stack, key commands. Add rules ONLY when Claude makes a mistake.

# Project: MyApp
# Stack: React 19 + TypeScript + Tailwind 4
# Commands: pnpm dev | pnpm test | pnpm lint
# Rules:
# - Use named exports only
# - Never use 'any' type
# - Run tests before committing

Root CLAUDE.md for global rules. Folder-level CLAUDE.md files for module-specific rules. Claude reads the closest one.

# Root: ./CLAUDE.md → global rules
# Module: ./src/api/CLAUDE.md → API-specific rules
# Module: ./src/ui/CLAUDE.md → UI-specific rules

~/.claude/settings.json for global. .claude/settings.json for project. Project overrides global.

// ~/.claude/settings.json (global)
{
  "permissions": { "allow": ["Read", "Write", "Bash(npm test)"] },
  "env": { "CLAUDE_CODE_MAX_OUTPUT_TOKENS": "16000" }
}

Skip all permission prompts for maximum speed. Use in trusted environments only. The 'YOLO mode'.

claude --dangerously-skip-permissions

Opt out of Anthropic's usage data collection for privacy.

claude config set --global preferredNotifChannel terminal
claude config set --global disableTelemetry true

Claude reads up to 150K characters of terminal output. Redirect large outputs to files to avoid truncation.

# Instead of: npm test (huge output)
# Do: npm test > test-results.txt 2>&1 && cat test-results.txt

Claude auto-compacts when context hits 75% capacity. You can also trigger manually with /compact.

Use different models for different tasks. Haiku for simple edits, Sonnet for most work, Opus for complex architecture.

/model sonnet
/model opus
/model haiku

Set instructions that apply to ALL your projects in ~/.claude/CLAUDE.md

# ~/.claude/CLAUDE.md
# Always use TypeScript strict mode
# Prefer functional components
# Write tests for all new functions

The #1 workflow for serious projects. Claude interviews you, creates a PRD, then plans and builds from it. Prevents the biggest mistake: building before thinking.

Create a detailed PRD for [describe your app]. Ask me at least 10 questions to understand the requirements before writing any code.

Always use Plan Mode first. Describe what you want. Review the plan. THEN switch to execute. Never let Claude code without a plan.

Shift+Tab (plan) → 'Build a user auth system with JWT' → review plan → Shift+Tab (execute)

Write tests first, then have Claude implement until they pass. Prevents hallucinated 'it works' claims.

Write failing tests for [feature]. Then implement the code to make all tests pass. Run tests after every change.