Reference

Claude Code Cheatsheet

Every command, shortcut, configuration option, and syntax pattern in one searchable reference. 96 items across 12 categories.

Core CLI Commands

claudeStart an interactive REPL session

claude -p "prompt"Run a one-shot prompt without interactive mode

claude -cContinue the most recent conversation

claude -r idResume a specific conversation by session ID

claude -wStart a session in a new git worktree (parallel work)

claude -nStart a new session explicitly (no resume prompt)

claude updateUpdate Claude Code to the latest version

claude doctorRun diagnostics to check your setup

claude mcp add nameAdd a new MCP server by name

claude mcp listList all configured MCP servers

claude mcp remove nameRemove an MCP server

CLI Flags

--model <name>Override the default model (e.g. claude-sonnet-4-20250514)

--effort <level>Set reasoning effort: low, medium, high, or max

--continueContinue the most recent conversation

--resume <id>Resume a specific conversation by session ID

--worktreeCreate a git worktree for parallel work

--max-turns <n>Limit the number of agentic turns in non-interactive mode

--output-format jsonOutput as JSON (for scripting/piping)

--verboseEnable verbose logging output

--add-dir <path>Add additional directory context

--allowedTools <tools>Restrict available tools (comma-separated list)

--from-pr <url>Start a session from a GitHub PR URL

Essential Slash Commands

/clearClear conversation history and start fresh

/compactSummarize and compact the conversation context

/contextShow the current context window usage

/model <name>Switch to a different model mid-session

/effort <level>Change reasoning effort level

/planEnter plan mode — Claude drafts a plan before acting

/executeExecute the current plan

/forkFork the conversation into a parallel branch

/rewindGo back to a previous message in the conversation

/resumeList and resume a previous conversation

/batchRun a prompt across multiple files

/simplifyReview changed code for reuse, quality, and efficiency

/loop <interval> <cmd>Run a slash command on a recurring interval

/voiceToggle voice input mode

/mcpManage MCP servers in-session

/hooksView and manage lifecycle hooks

/configView or modify configuration

/helpShow all available commands

Keyboard Shortcuts

Ctrl + CCancel current generation or tool execution

Ctrl + DExit the REPL / end session

TabAutocomplete file paths and commands

Alt + TToggle tool output visibility

Ctrl + OOpen a file in your editor from the session

Esc EscDouble-tap Escape to cancel and edit your last message

Shift + EnterInsert a newline in multi-line input

Ctrl + BBackground the current session

Shift + Up/DownScroll through message history

CLAUDE.md Locations

~/.claude/CLAUDE.mdUser-level — applies to all your sessions everywhere

./CLAUDE.mdProject root — shared with the team via git

./.claude/CLAUDE.mdProject root (hidden) — alternative location

./src/CLAUDE.mdSubdirectory — scoped rules for that folder and below

Auto-memory (~/.claude/user_memory.md)Claude learns preferences over time and saves them here

Configuration Files

~/.claude/settings.jsonUser-level settings — permissions, model prefs, global tool config

.claude/settings.jsonProject-level settings — shared with the team (committed)

.claude/settings.local.jsonProject-level local overrides — gitignored, personal tweaks

Hook Events

PreToolUseFires before Claude runs any tool — validate, block, or modify

PostToolUseFires after a tool completes — transform output, trigger side effects

UserPromptSubmitFires when the user sends a message — preprocess, validate input

SessionStartFires when a new session begins — setup environment, load context

NotificationFires when Claude sends a notification — route to Slack, email, etc.

TeammateIdleFires when an agent teammate goes idle — reassign work, alert user

TaskCompletedFires when an agent finishes a task — log results, chain next task

ElicitationFires when Claude asks the user a question — auto-answer, log

ElicitationResultFires when the user responds to an elicitation

WorktreeCreateFires when a git worktree is created — setup hooks, install deps

WorktreeRemoveFires when a git worktree is removed — cleanup

MCP Transports

stdioLocal process communication — best for local tools and scripts

HTTP (Streamable HTTP)Remote server — for cloud-hosted MCP servers

SSE (Server-Sent Events)Legacy streaming transport — still supported but prefer HTTP

Scope: user (~/.claude/)User-scoped MCP server — available in all your projects

Scope: project (.claude/)Project-scoped MCP server — shared with team

Effort Levels

lowQuick answers, minimal reasoning — for simple lookups and yes/no questions

mediumBalanced — default level, good for most tasks

highThorough reasoning — for complex code reviews, architecture decisions

maxMaximum reasoning depth — for the hardest problems, multi-step analysis

ultrathinkKeyword in prompt to trigger extended thinking budget beyond max

Environment Variables

ANTHROPIC_API_KEYYour API key — required for direct API usage

ENABLE_LSP_TOOLEnable the LSP (Language Server Protocol) tool for code intelligence

MAX_THINKING_TOKENSOverride the maximum tokens allocated to extended thinking

CLAUDE_CODE_EFFORT_LEVELSet default effort level via environment variable

ENABLE_SESSION_BACKGROUNDINGAllow backgrounding sessions with Ctrl+B

Permission Patterns

ReadAllow reading all files

Read(src/**)Allow reading only files under src/

WriteAllow writing to all files

Write(src/**)Allow writing only to files under src/

Bash(git *)Allow git commands only

Bash(npm test)Allow only npm test command

mcp__*Allow all MCP tools

mcp__server__toolAllow a specific MCP server tool

In-Session Syntax

@file.tsReference a specific file — adds it to context

@src/components/Reference a directory — adds file listing to context

@**/*.test.tsReference files via glob pattern

!npm testRun a shell command and include its output in context

@agent-nameMention a teammate agent by name in multi-agent setup