Claude Code — Cheat Sheet 📋

"The page you print and pin next to your screen"

SectionInstall & Setup

Claude Code is Anthropic's official terminal coding agent — it reads/edits files, runs shell commands, uses MCP servers, delegates to subagents, and can run headless.

# macOS / Linux / WSL — native install (recommended)
curl-fsSLhttps://claude.ai/install.sh|bash

# Windows PowerShell
irmhttps://claude.ai/install.ps1|iex

# Windows CMD
curl-fsSLhttps://claude.ai/install.cmd-oinstall.cmd&&install.cmd&&delinstall.cmd

# Stable channel instead of latest
curl-fsSLhttps://claude.ai/install.sh|bash-sstable

# A specific version
curl-fsSLhttps://claude.ai/install.sh|bash-s2.1.89

# Homebrew (no auto-update)
brewinstall--caskclaude-code

# WinGet
wingetinstallAnthropic.ClaudeCode

# npm (needs Node.js 22+ — installs the SAME native binary; claude does NOT use your Node at runtime)
npminstall-g@anthropic-ai/claude-code
# ⚠️ NEVER: sudo npm install -g

Auth: If the ANTHROPIC_API_KEY env var is set, Claude prompts once to approve the key instead of opening a browser. Enterprise providers: Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry.

System requirements: macOS 13+ / Windows 10 1809+ / Ubuntu 20.04+ / Debian 10+ / Alpine 3.19+, 4GB+ RAM, x64/ARM64, ripgrep bundled. On native Windows, Git for Windows is optional (enables the Bash tool via Git Bash; otherwise it uses the PowerShell tool).

SectionCore CLI Commands

SectionKey Flags

SectionHeadless & Programmatic

# One-shot: print and exit
claude-p"explain the root cause"

# Structured JSON output (result, cost, session_id...)
claude-p"summarize this diff"--output-formatjson

# Streaming NDJSON events — with streaming input
claude-p"refactor auth"--output-formatstream-json--input-formatstream-json

# Pipe content from another tool
caterror.log|claude-p"explain the root cause"

# Automation guardrails
claude-p"fix failing tests"\
--max-turns8\
--max-budget-usd2.00\
--allowedTools"Read,Edit,Bash(npm test:*)"\
--permission-modeacceptEdits\
--session-id123e4567-e89b-12d3-a456-426614174000

# Long-lived auth for CI
claudesetup-token

The Claude Agent SDK (TypeScript / Python) wraps all of this so you can build custom agents.

SectionPermission Modes

Switching: Shift+Tab cycles the modes (default → acceptEdits → plan → …). Or start directly in a mode with --permission-mode.

SectionSession & Conversation

SectionAll Slash Commands

Built-in

Bundled skills (ship with Claude Code)

Aliases: /settings=/config · /reset,/new=/clear · /cost=/usage · /bug,/share=/feedback · /quit=/exit · /checkup=/doctor · /allowed-tools=/permissions · /bg=/background · /app=/desktop · /proactive=/loop.

SectionDelegation & Subagents

Subagents are specialized AI assistants that handle a specific kind of task in their own context window, with their own system prompt, tool access, and permissions — and they return only a summary to the main chat.

Why they matter: (1) preserve the main context, (2) enforce tool constraints, (3) reuse config across projects, (4) specialize behavior, (5) control cost by routing tasks to cheap/fast models like Haiku.

How delegation happens: Claude reads each subagent's description and delegates automatically when a task matches. You can also force it:

>Usethecode-reviewersubagenttocheckmychanges

Managing them & file locations:

Frontmatter fields: name · description (when to use it — write it well!) · tools (optional; omit = inherit all) · model (optional: sonnet/opus/haiku or 'inherit').

---
name: code-reviewer
description: Expert code review specialist. Use PROACTIVELY after writing or changing code.
tools: Read, Grep, Glob, Bash
model: sonnet
---
You are a senior code reviewer. When invoked: run git diff, focus on changed files, review for
readability, bugs, security. Give feedback grouped by priority.

Mental model: the main agent = orchestrator; subagents = focused workers with clean context; only the summary comes back.

SectionCustom Commands (Skills)

Custom commands are merged into skills. .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both make /deploy.

Frontmatter fields: description · argument-hint · allowed-tools · model · (skills add) invocation control + run-in: subagent.

Body specials:

• $ARGUMENTS = everything typed after the command · $1, $2, … = positional args.
• ! at the start of a line = run a bash command and embed its output (needs allowed-tools Bash).
• @path/to/file = embed a file's contents.

Namespacing: subdirectories → /frontend:lint, /backend:deploy. Skills load their body only when used (cheap until needed). Claude can auto-invoke a skill when relevant, or you invoke /name.

---
description: Create a git commit
argument-hint: [message]
allowed-tools: Bash(git add:*), Bash(git commit:*)
model: claude-haiku-4-5
---
## Context
- Current status: !`git status`
- Current diff: !`git diff HEAD`
## Task
Create a single commit with message: $ARGUMENTS

SectionMemory & Config

CLAUDE.md = persistent instructions, auto-loaded. Hierarchy (all concatenate, later can add to earlier):

• Nested CLAUDE.md in subdirs load on demand when Claude touches files there.
• /init = scaffold a CLAUDE.md for the project.
• /memory = open + edit memory files.
• # at the start of input = "remember this" → appends a line to a memory file.
• @path/import.md inside CLAUDE.md = import another file (max 5 hops).
• Auto-memory: Claude writes notes to ~/.claude/projects/<project>/memory/ (MEMORY.md index). Survives /compact. Gitignored.
• Path-scoped rules: .claude/rules/*.md with paths: frontmatter globs load only when Claude reads matching files.

settings.json hierarchy (precedence high → low)

Managed(org)→command-lineflags→.claude/settings.local.json(personal,gitignored)
  → .claude/settings.json (project, shared) → ~/.claude/settings.json (user)

Key fields: model · permissions {allow, ask, deny, defaultMode} · env · hooks · autoUpdatesChannel · minimumVersion · editorMode ("vim"/"normal") · autoMemoryEnabled · outputStyle.

SectionHooks

Deterministic handlers (shell/HTTP/…) fired at lifecycle events (settings.json → hooks).

There are more: PermissionRequest, PermissionDenied, TaskCreated/Completed, FileChanged, WorktreeCreate/Remove, etc.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "if": "Bash(rm *)",
            "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/block-rm.sh",
            "args": []
          }
        ]
      }
    ]
  }
}

• Matcher: ""/"*"/omitted = all · plain string = exact (e.g. "Edit|Write") · other chars = regex.
• Hook types: command, http, mcp_tool, prompt, agent. Common fields: if, timeout (default 600s), statusMessage.
• Exit codes: 0 = ok (stdout may carry JSON) · 2 = BLOCK (stderr → Claude) · other = non-blocking error.
• JSON control (PreToolUse): {"hookSpecificOutput":{"hookEventName":"PreToolUse","permissionDecision":"deny","permissionDecisionReason":"..."}} — values allow/deny/ask/defer.
• Universal JSON: continue (false stops Claude) · stopReason · suppressOutput · systemMessage · additionalContext.
• Path vars: ${CLAUDE_PROJECT_DIR} · ${CLAUDE_PLUGIN_ROOT}.

SectionMCP

Connect external tools/data.

# stdio
claudemcpadd<name>--<command>[args]

# SSE / HTTP
claudemcpadd--transportsse<name><url>
claudemcpadd--transporthttp<name><url>

# From JSON
claudemcpadd-json<name>'<json>'

# Import from Claude Desktop
claudemcpadd-from-claude-desktop

# List / details / remove
claudemcplist
claudemcpget<name>
claudemcpremove<name>

• In the REPL: /mcp shows servers + status, does OAuth login (/mcp → authenticate), reconnect/enable/disable.
• Use resources in prompts: @server:protocol://resource.
• MCP servers can expose prompts as slash commands: /mcp__server__promptname.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "..." }
    }
  }
}

SectionPermission Rules

allow/ask/deny rules in settings.json (permissions.allow/ask/deny):

/permissions opens the rules UI. defaultMode in settings sets the starting mode.

SectionShortcuts & Input

Prefixes at the start of a line:

Image input: drag/paste screenshots for Claude to analyze.

SectionEnv Vars & Config Paths

SectionQuick Comparison