Master Claude Code: Install, Shortcuts, and Advanced Workflows

Installation

Claude Code can be installed on macOS/Linux, Homebrew, Windows PowerShell, or via npm (Node 18+): curl -fsSL https://claude.ai/install.sh – macOS/Linux brew install --cask claude-code – Homebrew irm https://claude.ai/install.ps1 – Windows PowerShell npm install -g @anthropic-ai/claude-code – npm

Basic CLI Commands

cd your-project && claude

– start an interactive session in a project claude update – update to the latest version claude -c – continue the previous conversation claude "prompt" – start a session with an initial prompt claude -r <session-id> – resume a specific session claude -t – enable Git worktree mode

Shortcut Keys (interactive UI)

Enter – send / submit

Esc – interrupt generation

Esc Esc – open rewind UI

Ctrl C – open main menu

Ctrl R – open external editor / search history

Ctrl E – exit Claude Code

Shift Tab – toggle Auto‑Accept mode (cycle with Shift Tab Tab for Plan mode)

Ctrl L – toggle tool list

Ctrl \ – toggle side panel

Ctrl J / Ctrl V – paste image

Ctrl K – show prompt or delete after cursor

Ctrl A / Ctrl E – move to line start / end

Opt F / Opt B – word forward / backward

↑ / ↓ – browse command history

Shift F – model selector

Shift T – extend thinking

Ctrl Tab – run background task

Slash Commands

/clear

– clear conversation history /compact – compress token usage /resume – resume a paused conversation /reset or /rewind – rollback to a previous code node or conversation state /cost – display token usage and estimated cost /rate-limits – show current rate limits /convert – visualise token stream /config – open settings panel /permissions – manage permission mode /vim – toggle Vim editing mode /init – create CLAUDE.md in the project /memory – edit the memory file /review – run code review /doctor – perform environment diagnostics /agents – list or manage sub‑agents /mcp – manage MCP (external tool) servers /tasks – list pending tasks

Permission Modes

Normal – each action requires confirmation (most secure)

Auto‑Accept – actions are executed automatically (fast)

Plan – only generate a plan; no code execution

Switch modes with Shift Tab (cycle: Normal → Auto‑Accept → Plan).

CLI Startup Parameters

--model sonnet|haiku|opus

– select the Anthropic model --agent <name> – invoke a specific sub‑agent --permission-mode plan|auto|normal – set initial permission mode --max-turns N – limit the number of interaction turns --output-format text|json|stream-json – choose output serialization --allowedTools <list> – whitelist tools --disallowedTools <list> – blacklist tools --verbose – enable detailed logging --no-browser – disable automatic browser integration

Headless (non‑interactive) Mode

claude -p "query"

– output result directly to stdout echo "fix bug" | claude -p --output-format json – pipe input and receive JSON response

Extensions System

Project‑level memory files: ~/CLAUDE.md or .claude/CLAUDE.md Global memory file: ~/.claude/CLAUDE.md Project slash‑commands directory: .claude/commands/ Global slash‑commands directory: ~/.claude/commands/ Built‑in tools: Bash, Read, Write (auto‑invoked)

Hook system: event‑driven automation (e.g., PostToolUse)

MCP server: external tool extensions

Sub‑Agents and Skills

Agents stored in .claude/agents/ (project) or ~/.claude/agents/ (global)

Skills stored in .claude/skills/ (project) or ~/.claude/skills/ (global)

MCP Server Management

claude mcp add <url>

– add a server claude mcp list – list configured servers claude mcp remove <url> – remove a server

Git Worktree Integration

git worktree add <path> <branch>

– create a new worktree for parallel development git worktree list – list existing worktrees git worktree remove <path> – delete a worktree

Use worktrees to develop features, hot‑fixes, or experiments without affecting the main branch.

Typical Project Workflow

Navigate to the project directory and start Claude Code: cd my-project && claude Initialize the project memory file: /init Press Shift Tab to enter Plan mode, review the generated plan, then switch to execution (Auto‑Accept or Normal mode).

Use slash commands (e.g., /review, /doctor) and shortcuts to iterate quickly.

Efficient Debugging

Paste an error stack trace; Claude Code will locate the problematic code.

Command example: cat error.log | claude -p "fix" returns a suggested fix.

Run background tasks with Ctrl Tab while continuing the main conversation.

Hook Configuration Example (JSON)

{
  "PostToolUse": [{
    "matcher": "Edit|Write",
    "command": "npx prettier --write $FILE_PATH"
  }]
}

Configuration Files (priority order)

/etc/claude-code/managed-settings.json

.claude/settings.json

.claude/settings.local.json

~/.claude/settings.json

Rewind Options

Conversation Only – rollback only the dialogue.

Code – rollback only generated code.

Full Rewind – rollback both conversation and code.

Note: Checkpoints do not track external Bash changes; use Git for full traceability.