Claude Code Cheatsheet

Install on macOS/Linux (auto-updates). No Node.js required.

Install via Homebrew (manual updates).

Install on Windows PowerShell.

Check the installed version.

Health check: verify install, network, and config (inside a Claude Code session).

Update to the latest version.

Launch → sign in for the first time via browser.

Log in or switch CLI accounts.

Basic plan, enough for individual use (usage limits apply).

5x the usage of Pro (for devs working all day).

20x the usage of Pro (small teams/power users).

Pay per token (~$100-200/dev/month). Use --max-budget-usd to set a limit.

Centralized management, SSO, team billing.

Prefer subscription plans for predictable billing; use the API only for CI/CD or headless workflows.

Compact context: the most important command (do not wait until context runs out).

Clear everything, start a new session.

Enter Plan Mode: Claude only reads, it does not edit code.

Switch models (haiku/sonnet/opus/opusplan).

Enable fast mode.

Initialize a project: auto-generate CLAUDE.md.

View token cost (API users).

Analyze your work session: patterns, friction points.

Open settings: theme, model, output style.

Switch permission mode (Default/Auto/Plan).

Undo: go back to the previous step.

Toggle thinking: deeper reasoning.

Asks before editing files or running commands (safest).

Auto-edits files, asks before running commands.

Read only: no edits, no execution.

Skips everything: only use inside a container/VM (DANGEROUS).

Quickly switch between modes.

Choose the mode when launching the CLI.

Attach a file: Claude reads and understands it.

Quickly run a shell command.

Quick question that does not affect context.

Quick headless question: use in scripts/CI/CD.

Continue the previous session.

Return to a specific session by ID.

Choose a model at startup.

Add a directory outside the project.

Pull a session from the web into the terminal.

Output JSON (text/json/stream-json).

Limit the number of agent turns (headless).

Limit cost per session.

Project memory (repo root or .claude/). Commit to git.

Personal memory: only you can see it.

Auto-generate CLAUDE.md for the project.

Hooks and permissions for the project (commit this).

Personal permissions (gitignore this).

Skill modules: in-depth expertise.

Rules automatically loaded into context.

Detects over-engineering and auto-fixes it.

Runs a command repeatedly at an interval.

Name a session so it is easy to find again.

Select and copy a code block from the conversation.

View usage graph, streaks, and favorite model.

Pick and return to an older session.

Esc×2 to rewind → creates a parallel timeline.

Sub-agents run in the background while you keep coding.

Scripts that auto-run on events (SessionStart, PreToolUse...).

Check login status.

Add an MCP server (e.g., context7).

Comfortable: safe zone.

Careful: run /compact soon.

URGENT: /compact or /clear right now.

Small fixes, review: cheapest.

Everyday coding: best value.

Hard debugging, architecture: most powerful.

Enable when you need deeper analysis.

The concrete deliverable to create.

A clear file path.

Constraints, tech, approach.

Pass/fail criteria.

Switch permission mode.

Undo / rewind the previous step.

Toggle thinking.

Stop Claude while it is processing.

Search command history.

Clear the screen (keeps context).

View background tasks.

Cancel all agents.

Autocomplete.

New line (does not send).

Exit.

Open Claude Code in VS Code.

Open Claude Code in JetBrains.

View and manage connected MCP servers.

Debug mode: view detailed logs at startup.

MCP for library documentation.

MCP for browser automation.

MCP for database queries.