📖 Quick Reference

The commands and settings you'll look up most often when using Claude Code, all in one place.

---

Basic Usage

claude                    # Start interactive mode
claude "your task"        # Run a single prompt
claude -p "prompt"        # --print mode (output only)
claude --continue         # Continue the last conversation
claude --resume           # Pick a session and resume
claude --model sonnet     # Specify model (sonnet, opus, fable, best, or full model name)
claude update             # Update to the latest version
claude auth login         # Log in with your Anthropic account
claude auth status        # Check auth status (JSON)
claude agents             # List configured subagents
claude mcp                # Configure MCP servers

---

Slash Commands

Command	Description
/help	Show help
/clear	Reset the current conversation (aliases: /reset, /new)
/compact [instructions]	Summarize the conversation to save context (optional summary instructions)
/memory	Edit CLAUDE.md files, configure auto-memory
/cost	Check cumulative cost for the current session
/model [model]	Change the active model (adjust Effort Level with left/right arrows)
/permissions	Review tool permission settings (alias: /allowed-tools)
/status	Check version, model, account, and connection status
/config	Open the settings interface (alias: /settings)
/init	Generate an initial CLAUDE.md
/fast [on|off]	Toggle Fast Mode
/diff	Interactive diff viewer for uncommitted changes
/context	Visualize current context usage
/copy	Copy the last response to the clipboard
/doctor	Diagnose installation and configuration
/export [filename]	Export the current conversation as text
/fork [name]	Copy the conversation into a new background session (shown in claude agents, v2.1.212+) — for in-session branching use /subtask
/hooks	Manage hook configuration
/plugin	Manage plugins
/pr-comments [PR]	Fetch GitHub PR comments
/rename [name]	Rename the current session
/resume [session]	Resume a session (alias: /continue)
/rewind	Roll back to an earlier point (alias: /checkpoint)
/sandbox	Toggle sandbox mode
/skills	List available Skills
/stats	Visualize daily usage and session history
/theme	Change the color theme
/vim	Switch between Vim/Normal editing modes

tip

/review <pr> is a fast single-pass review; for multi-agent review, run /code-review <level> <pr#> and pick the intensity (v2.1.202+). As of v2.1.215, /verify and /code-review are no longer run automatically — invoke them yourself when needed.

---

Key CLI Flags

Flag	Description
--print, -p	Output only (non-interactive mode)
--continue, -c	Continue the most recent conversation in the current directory
--resume, -r	Resume by session ID or name
--model	Specify model (sonnet, opus, fable, best, or full model name)
--permission-mode	Specify permission mode (default, plan, acceptEdits, bypassPermissions)
--allowedTools	Tools to auto-approve
--disallowedTools	Tools to prohibit
--tools	Restrict available tools ("Bash,Edit,Read")
--add-dir	Add extra working directories
--system-prompt	Replace the entire system prompt
--append-system-prompt	Append instructions to the default prompt
--max-turns	Max turns for the agentic loop (print mode)
--max-budget-usd	API cost cap (print mode)
--output-format	Output format (text, json, stream-json)
--mcp-config	Load an MCP server config JSON file
--worktree, -w	Run in an isolated git worktree
--chrome / --no-chrome	Enable/disable Chrome browser integration
--agent	Specify the agent for the session
--agents	Custom subagent JSON definitions
--plugin-dir	Load a plugin directory
--dangerously-skip-permissions	Skip all permission checks (caution!)
--verbose	Verbose logging (per-turn output)
--debug	Debug mode (category filters supported)
--json-schema	Structured output conforming to a JSON Schema (print mode)
--fallback-model	Fallback model when the primary is overloaded (print mode)
--fork-session	Fork into a new session ID when resuming
--from-pr	Resume the session linked to a GitHub PR
--remote	Create a web session on claude.ai
--teleport	Bring a web session into your local terminal
--teammate-mode	Agent Team display mode (auto, in-process, tmux)

tip

--append-system-prompt is recommended in most cases because it keeps the default Claude Code behavior and only adds your instructions. --system-prompt replaces everything, so use it with care.

---

Permission Modes

Switch with Shift+Tab, or set at startup with the --permission-mode flag:

Mode	Value	Description
Default	default	Ask for approval on each action
Auto-accept Edits	acceptEdits	Auto-approve file edits; Bash still requires confirmation
Plan Mode	plan	Read-only exploration, no file modifications
Bypass	bypassPermissions	Skip all permission checks

caution

Use --dangerously-skip-permissions (Bypass mode) only in isolated environments (Docker, VMs).

---

CLAUDE.md Template

# Project Name

## Tech Stack
- Language: TypeScript
- Framework: Next.js
- Package manager: npm

## Coding Conventions
- Function names: camelCase
- File names: kebab-case
- Indentation: 2 spaces

## Common Commands
- Dev server: `npm run dev`
- Build: `npm run build`
- Test: `npm test`

## Current Progress
- ✅ Completed work
- 🔄 In progress
- ⬜ Planned

---

Environment Variables

Core Settings

Variable	Description	Default
ANTHROPIC_API_KEY	API authentication key	—
ANTHROPIC_MODEL	Default model setting	Varies by plan
CLAUDE_CODE_MAX_OUTPUT_TOKENS	Maximum output tokens	32,000
CLAUDE_CODE_EFFORT_LEVEL	Effort Level (low, medium, high)	—
CLAUDE_CODE_DISABLE_FAST_MODE	Disable Fast Mode (1)	—
CLAUDE_CODE_SHELL	Override automatic shell detection	—
CLAUDE_CONFIG_DIR	Change where config/data files are stored	—
HTTP_PROXY / HTTPS_PROXY	Proxy settings	—

Feature Toggles

Variable	Description
CLAUDE_CODE_DISABLE_AUTO_MEMORY	Disable auto-memory (1) / force enable (0)
CLAUDE_CODE_DISABLE_BACKGROUND_TASKS	Disable background tasks (1)
CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS	Remove built-in commit/PR workflow instructions (1)
CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING	Disable adaptive thinking (1) — applies to Opus 4.6 and Sonnet 4.6 only; Opus 4.7+ is always adaptive
CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION	Disable prompt suggestions (false)
CLAUDE_CODE_ENABLE_TASKS	Disable task tracking, restoring the previous TODO list (false)
DISABLE_AUTOUPDATER	Disable auto-updates (1)
DISABLE_PROMPT_CACHING	Disable prompt caching (1)
DISABLE_COST_WARNINGS	Disable cost warning messages (1)
DISABLE_TELEMETRY	Disable telemetry (1)

Bash Tool Settings

Variable	Description
BASH_DEFAULT_TIMEOUT_MS	Default timeout for Bash commands
BASH_MAX_TIMEOUT_MS	Maximum timeout for Bash commands
BASH_MAX_OUTPUT_LENGTH	Maximum Bash output length in characters (middle is truncated beyond this)

Cloud Providers (Bedrock / Vertex / Foundry)

Variable	Description
CLAUDE_CODE_USE_BEDROCK	Use AWS Bedrock
CLAUDE_CODE_USE_VERTEX	Use Google Vertex
CLAUDE_CODE_USE_FOUNDRY	Use Microsoft Foundry
ANTHROPIC_FOUNDRY_BASE_URL	Foundry resource URL

tip

For the full list of environment variables, see the official settings documentation.

---

Model Selection Guide

Model	Use Case	Relative Cost
claude-haiku-4-5	Fast tasks, bulk processing	Cheap
claude-sonnet-5	General development (recommended default, released 2026-06-30)	Medium
claude-opus-4-8	Complex design, deep reasoning	Expensive
claude-fable-5	The hardest reasoning, long autonomous runs (1M default, v2.1.170+)	Very expensive

---

Hook Event Types (18)

Event	When It Fires
SessionStart	When a session starts or resumes
UserPromptSubmit	After a prompt is submitted, before Claude processes it
PreToolUse	Right before a tool runs (can block)
PermissionRequest	When a permission dialog is shown
PostToolUse	After a tool runs successfully
PostToolUseFailure	After a tool run fails
Notification	When Claude sends a notification
SubagentStart	When a subagent is created
SubagentStop	After a subagent finishes
Stop	After Claude finishes responding
TeammateIdle	When an Agent Team teammate goes idle
TaskCompleted	When a task is marked complete
InstructionsLoaded	When CLAUDE.md or rules files are loaded
ConfigChange	When a config file changes mid-session
WorktreeCreate	When a worktree is created
WorktreeRemove	When a worktree is removed
PreCompact	Before context compaction
SessionEnd	When a session ends

tip

Hooks support four types: shell command (command), HTTP endpoint, LLM prompt, and MCP tool. For configuration details, see the official Hooks reference.

---

Agent SDK Quick Reference

import { query } from "@anthropic-ai/claude-agent-sdk";

// Basic run
const response = await query({
  prompt: "Your task",
  options: {
    maxTurns: 10,
    allowedTools: ["Read", "Write", "Bash"],
    systemPrompt: "You are an expert."
  }
});

// Extract response text
const text = response
  .filter(m => m.type === "assistant")
  .flatMap(m => m.content)
  .filter(b => b.type === "text")
  .map(b => b.text)
  .join("\n");

---

Common Prompt Patterns

# Code review
"Review the changes in this PR and point out bugs, security issues, and improvements."

# Refactoring
"Refactor this function for readability. Don't change its behavior."

# Writing tests
"Write unit tests for this function with Jest. Include edge cases."

# Documentation
"Write JSDoc comments for this module."

# Debugging
"Find the cause of this error and fix it: [error message]"