Reference · module 4 of 6

Claude Code cheat sheet

Verify against the CLI reference — this ships weekly.

Install

curl -fsSL https://claude.ai/install.sh | bash     # macOS / Linux / WSL
irm https://claude.ai/install.ps1 | iex             # Windows PowerShell
brew install --cask claude-code                     # stable channel
brew install --cask claude-code@latest              # latest channel
winget install Anthropic.ClaudeCode
claude --version

Native installs auto-update. Homebrew and WinGet do not.

Shell commands

Command	What
`claude`	Interactive session
`claude "task"`	Interactive, with an initial prompt
`claude -p "query"`	Print and exit
`cat f \\| claude -p "q"`	Pipe content in
`claude -c`	Continue the most recent here
`claude -r [id\\|name]`	Resume — picker or direct
`claude -n "name"`	Name the session
`claude update`	Update
`claude install [version]`	Install a specific version
`claude auth login\\|logout\\|status`	Auth
`claude setup-token`	Long-lived OAuth token for CI
`claude doctor`	Read-only diagnostics from the shell
`claude mcp`	Manage MCP servers
`claude mcp login\\|logout <name>`	MCP OAuth from the shell
`claude plugin`	Manage plugins
`claude agents`	Agent view dashboard
`claude attach\\|logs\\|stop\\|respawn\\|rm <id>`	Background session control
`claude daemon status\\|stop`	Supervisor control
`claude project purge [path]`	Delete local state for a project
`claude remote-control`	Remote Control server
`claude ultrareview [target]`	Non-interactive deep review
`claude gateway --config gateway.yaml`	Self-hosted gateway

Key flags

Model and effort

--model sonnet|opus|haiku|fable|<full-id>
--effort low|medium|high|xhigh|max|ultracode
--fallback-model sonnet,haiku
--advisor opus|sonnet

Permissions and tools

--permission-mode default|acceptEdits|plan|auto|dontAsk|bypassPermissions|manual
--allowedTools "Bash(git log *)" "Read"
--disallowedTools "Edit" "mcp__*"
--tools "Bash,Edit,Read"
--dangerously-skip-permissions
--allow-dangerously-skip-permissions

Print mode

-p / --print
--output-format text|json|stream-json
--input-format text|stream-json
--json-schema '{...}'
--max-turns N
--max-budget-usd N
--verbose
--include-partial-messages
--forward-subagent-text
--no-session-persistence

System prompt

--append-system-prompt "text"
--append-system-prompt-file ./file
--system-prompt "text"           # replaces everything
--system-prompt-file ./file      # replaces everything
--append-subagent-system-prompt "text"
--exclude-dynamic-system-prompt-sections

Config and startup

--add-dir ../lib ../apps
--settings ./settings.json
--setting-sources user,project,local
--mcp-config ./mcp.json
--strict-mcp-config
--plugin-dir ./plugin
--plugin-url https://...
--agents '{"name":{"description":"...","prompt":"..."}}'
--agent my-agent
--bare                            # skip all auto-discovery; fast
--safe-mode                       # all customisations disabled
--debug "api,mcp"
--debug-file /tmp/claude.log

Sessions and surfaces

-w, --worktree [name|#PR]
--tmux
--bg, --background
--exec 'pytest -x'
--fork-session
--session-id <uuid>
--from-pr 123
--cloud "task"
--teleport
--remote-control ["name"]
--ide
--chrome / --no-chrome
--teammate-mode in-process|auto|tmux|iterm2

Session commands

Command	What
`/help`	Everything
`/init`	Generate CLAUDE.md
`/context`	What’s loaded and its cost
`/memory`	View and edit memory files
`/usage`	What’s driving plan limits
`/doctor`	Diagnose config; propose CLAUDE.md trims
`/config`	Set any setting
`/hooks`	Browse registered hooks (read-only)
`/mcp`	Server status and token cost
`/plugin list`	Installed plugins
`/clear`	Clear history
`/compact`	Compact manually
`/rewind`	Resume from before a `/clear`
`/resume` `/rename`	Session management
`/model`	Change model
`/login` `/logout`	Auth
`/desktop`	Continue in the Desktop app
`/cd`	Move to a new directory
`/schedule`	Create a routine
`/goal`	Set a completion condition
`/loop`	Repeat a prompt
`/debug`	Enable debug logging
`/code-review [ultra]`	Review the working diff
`/review`	Review a GitHub PR
`/security-review`	Security review of pending changes
`/exit`	Quit

Keys: Shift+Tab cycles permission modes · Ctrl+O transcript view · Ctrl+R search command history across projects · ↑ history · Tab completes

File layout

project/
├── CLAUDE.md                 # or .claude/CLAUDE.md
├── CLAUDE.local.md           # personal, gitignore it
├── .mcp.json                 # project MCP servers
└── .claude/
    ├── settings.json         # committed
    ├── settings.local.json   # gitignored
    ├── rules/*.md            # modular, optionally path-scoped
    ├── skills/<name>/SKILL.md
    ├── agents/<name>.md
    ├── commands/*.md         # legacy; still works
    └── hooks/*.sh

~/.claude/
├── CLAUDE.md
├── settings.json
├── rules/ skills/ agents/
└── projects/<project>/memory/MEMORY.md

Precedence: managed policy > --settings > local > project > user

CLAUDE.md files: additive, all levels
Skills: managed > user > project
Subagents: managed > CLI flag > project > user > plugin
MCP: local > project > user
Hooks: all merge

Settings essentials

{
  "model": "sonnet",
  "fallbackModel": "haiku",
  "effortLevel": "medium",
  "autoMemoryEnabled": true,
  "sandbox": { "enabled": true },
  "permissions": {
    "deny": ["Read(.env)", "Read(**/*.pem)", "Bash(rm -rf /*)", "Bash(git push --force*)"],
    "ask":  ["Bash(git push*)", "Bash(npm publish*)"],
    "allow":["Read", "Grep", "Glob", "Bash(git status)", "Bash(git diff*)", "Bash(npm test*)"],
    "additionalDirectories": ["../shared-lib"]
  },
  "claudeMdExcludes": ["**/other-team/CLAUDE.md"],
  "hooks": { }
}

Hook events

SessionStart Setup UserPromptSubmit UserPromptExpansion PreToolUse PermissionRequest PermissionDenied PostToolUse PostToolUseFailure PostToolBatch Notification MessageDisplay SubagentStart SubagentStop TaskCreated TaskCompleted Stop StopFailure TeammateIdle InstructionsLoaded ConfigChange CwdChanged FileChanged WorktreeCreate WorktreeRemove PreCompact PostCompact Elicitation ElicitationResult SessionEnd

Types: command http mcp_tool prompt agent

Exit codes: 0 no objection (stdout enters context for UserPromptSubmit / UserPromptExpansion / SessionStart) · 2 block, stderr becomes Claude’s feedback · other = proceed with an error notice

PreToolUse hooks fire in every permission mode including bypass. Hooks tighten; they never loosen.

Skill frontmatter

---
name: my-skill                      # lowercase, hyphens, ≤64 chars
description: What it does. Use when [triggers].   # ≤1024 chars
disable-model-invocation: true      # only you can invoke; zero context cost
context: fork                       # run in an isolated subagent
allowed-tools: Read, Grep
model: haiku
---

Subagent frontmatter

---
name: test-auditor
description: When Claude should use it
tools: Read, Grep, Glob, Bash
model: sonnet
skills: [testing-conventions]
memory: true
---

Path-scoped rule

---
paths:
  - "src/api/**/*.ts"
  - "tests/**/*.{ts,tsx}"
---

Diagnostics, in order

/context      what actually loaded
/doctor       is the config broken
/hooks        which hooks are registered
/mcp          server status and cost
claude --safe-mode
claude --debug-file /tmp/claude.log