Some links in this article are affiliate links. We earn a commission at no extra cost to you when you purchase through them. Full disclosure.

Quick reference for Claude Code — Anthropic’s terminal AI coding agent.

Install & setup

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

# Start in your project
cd your-project
claude

# With a specific prompt
claude "Explain this codebase"

Slash commands

Command	What it does
/help	Show all commands
/compact [focus]	Summarize context, keep going. Add focus: /compact focus on auth refactor
/clear	Clear all context, start fresh
/rewind	Go back to a previous message (also: double-tap Esc)
/model	Switch model (Sonnet, Opus, Haiku)
/cost	Show token usage and cost for this session
/tokens	Show current context size
/config	Open settings
/permissions	View/edit permission settings
/bug	Report a bug to Anthropic
/ultrareview	🆕 Dedicated review session — reads changes, flags bugs and design issues a careful reviewer would catch. Pro/Max get 3 free.
/quit	Exit (also: Ctrl+C twice)

Keyboard shortcuts

Shortcut	Action
Esc Esc (double-tap)	Rewind to previous message
Ctrl+C	Cancel current generation
Ctrl+C Ctrl+C	Exit Claude Code
Tab	Accept suggestion
Enter	Send message
Shift+Enter	New line in message

Context management

# Check how much context you're using
/tokens

# Summarize and compress (best before context gets too long)
/compact

# Summarize with focus
/compact Focus on the database migration. Drop the CSS debugging.

# Nuclear option: start fresh with a handoff note
/clear
# Then paste: "We're refactoring auth. Files: src/auth/. Approach: JWT. Ruled out: OAuth2."

When to use each:

• /rewind — undo a bad attempt, try a different approach
• /compact — session getting long, keep working on same task
• /clear — switching to a completely different task

See our long-running agents guide for advanced context strategies.

Permission settings

// .claude/settings.json
{
  "permissions": {
    "allow": ["read_file", "list_directory", "search_files"],
    "ask": ["write_file", "shell_command"],
    "deny": ["delete_file"]
  }
}

Level	Behavior
allow	Auto-approved, no prompt
ask	Claude asks before each action
deny	Blocked, Claude can’t do it

Fix permission issues: Claude Code Permission Denied Fix

Subagents

# Tell Claude to use a subagent for isolated work
"Spin up a subagent to analyze the test coverage"
"Use a subagent to read the API docs and summarize the auth endpoints"

Subagents get their own context window. Only the result comes back. Use for: research tasks, file analysis, anything where you need the conclusion but not the intermediate steps.

Routines (cloud automation)

Set up at claude.ai/code/routines:

Trigger	Example
Schedule	”Every weekday at 9am: review new PRs”
API call	POST to trigger post-deployment checks
GitHub event	On PR opened: automated code review

See our Claude Code Routines guide for setup.

Model selection

Model	Best for	Speed	Cost
Sonnet	Daily coding (default on free)	Fast	$$
Opus 4.8	Complex architecture, hard bugs, agentic tasks, dynamic workflows	Slow	$$$$
Haiku	Quick questions, formatting	Fastest	$

/model sonnet   # Switch to Sonnet
/model opus     # Switch to Opus for hard problems
/model haiku    # Switch to Haiku for quick tasks

Effort levels 🆕

Opus 4.8 supports effort levels from low to max, with high as the default. Use ultracode to enable dynamic workflows (hundreds of parallel subagents for codebase-scale tasks).

Level	When to use	Token usage
low	Simple questions, formatting	Minimal
medium	Standard coding tasks	Moderate
high	Complex multi-file changes	High
xhigh	Hard bugs, architecture decisions (default)	High
max	Absolute best result, latency doesn’t matter	Very high

Set effort via the API effort parameter or let Claude Code manage it automatically.

Use with other models via API

# Use GLM-5.1 through Z.ai
export ANTHROPIC_BASE_URL="https://api.z.ai/api/anthropic"
export ANTHROPIC_AUTH_TOKEN="your-zai-key"
claude

See our GLM-5.1 + Claude Code setup guide.

Common patterns

# Code review
claude "Review src/auth/ for security issues"

# Refactoring
claude "Refactor the payment module to use the strategy pattern"

# Debugging
claude "The tests in src/auth/__tests__/ are failing. Fix them."

# Documentation
claude "Add JSDoc comments to all exported functions in src/lib/"

# Git workflow
claude "Create a PR description for the changes in this branch"

Pricing

Plan	Price	Includes
Pro	$20/mo	Sonnet + Haiku, generous limits
Max	$100/mo	Opus + Sonnet + Haiku, 5x limits
API	Pay-per-token	Full control, no subscription

Speed up your workflow — Raycast lets you trigger Claude Code from a keyboard shortcut.

Related: How to Use Claude Code · Claude Code Routines Guide · Claude Code Desktop App · What Is Claude Dispatch? · What Is Claude Cowork? · Claude Code vs Codex CLI vs Gemini CLI · GLM-5.1 + Claude Code Setup · Claude Code Permission Denied Fix