Claude Code Cheat Sheet: Complete Keyboard Shortcuts & Commands

Key Takeaways for Engineers

Terminal-first agentic coding CLI tools like Claude Code from Anthropic represent a significant productivity leap for modern software engineers, DevOps teams, and automation architects. By operating directly inside your terminal, Claude Code eliminates the constant context-switching between IDE windows, browser tabs, and shell environments.

This cheat sheet compiles every keyboard shortcut, readline navigation trick, multiline technique, slash command, and runtime CLI flag into a single reference guide designed for speed and daily terminal productivity.

Part 1: Interactive Mode Keyboard Shortcuts

Mastering terminal navigation inside an active Claude Code session enables rapid iteration without touching your mouse.

Session and Navigation Controls

• Ctrl + C: Cancels current execution, aborts long-running agent tool calls, or interrupts code generation safely without closing the session.
• Ctrl + D: Closes session cleanly if input buffer is empty (standard EOF signal).
• Ctrl + L: Clears the terminal screen while preserving the underlying conversation context and agent memory.
• Up / Down Arrows: Cycles backward and forward through prompt command history.
• Tab: Triggers contextual auto-completion for slash commands, local file paths, and git branch targets.

Readline-Style Text Editing Shortcuts

• Ctrl + A: Jump cursor instantly to the beginning of the line.
• Ctrl + E: Jump cursor directly to the end of the line.
• Ctrl + U: Clear the entire line before the current cursor position.
• Ctrl + K: Kill/delete all text from the cursor to the end of the line.
• Ctrl + W: Delete the preceding word behind the cursor.
• Alt + B (or Option + B): Move cursor backward by one full word.
• Alt + F (or Option + F): Move cursor forward by one full word.

Multiline Input Methods

• Shift + Enter: Inserts a newline without submitting the prompt (supported in modern terminal emulators).
• Ctrl + J: Standard terminal newline insertion fallback when Shift+Enter sends a carriage return.
• """ (Triple Quotes): Enclose code blocks, stack traces, or configuration files in triple quotes to enter multi-line prompt capture mode safely.
• Backslash (\) at line end: Escapes newline and continues prompt entry on the subsequent line.

Quick Input Prefixes

• !<command>: Direct bash command execution bypass (e.g. !git status or !npm test) executed locally without agent LLM reasoning overhead.
• @<path>: Explicitly attaches file contents or directory context into the immediate reasoning window.
• /<command>: Invokes specialized Claude Code runtime slash commands.

Part 2: Transcript Viewer and History Search

Navigating historical execution logs and multi-step agent tool executions requires dedicated overlay controls.

/btw Side Question Overlay

• /btw <question>: Opens a lightweight overlay to ask clarifying technical questions without polluting or appending tokens to your primary conversation context window.
• Esc: Closes the /btw overlay instantly and returns focus to your active prompt buffer.

History Search (Ctrl + R Mode)

• Ctrl + R: Activates reverse incremental search across historical prompts and submitted instructions.
• Enter: Selects the highlighted historical query and loads it into your active input line.
• Ctrl + G (or Esc): Cancels search and restores original input line.

Part 3: Vim Mode Navigation

Engineers accustomed to modal text editing can enable Vim keybindings within Claude Code settings for seamless buffer manipulation.

Mode Switching

• Esc: Switch from INSERT mode to NORMAL command mode.
• i: Enter INSERT mode before the cursor position.
• a: Enter INSERT mode after the cursor position.
• A: Jump to the end of the line and enter INSERT mode.
• I: Jump to the start of the first non-blank character and enter INSERT mode.

Navigation & Editing in NORMAL Mode

• h / j / k / l: Standard directional cursor movement (left, down, up, right).
• w / b / e: Jump forward by word, backward by word, or to word ending.
• 0 / $: Jump to absolute beginning or end of line.
• dd: Delete entire current line and save to register.
• dw: Delete word from cursor position forward.
• cw: Change word (deletes word and enters INSERT mode immediately).
• u: Undo previous in-buffer text modification.
• p: Paste register contents directly after cursor.

Part 4: Complete Slash Commands Reference

Slash commands control session context, invoke background workers, and govern agent capabilities.

Session & Context Management

• /compact: Summarizes and prunes conversation history to reclaim context window space while preserving key architectural decisions.
• /clear: Completely wipes conversation history and starts a fresh context window.
• /cost: Displays detailed breakdown of input tokens, output tokens, cache hits, and estimated session expenses.
• /context: Visualizes loaded files, memory files, system instructions, and token budget consumption.

Planning & Execution Modes

• /plan: Toggles step-by-step architectural planning mode before executing file modifications or bash operations.
• /review: Performs automated code review against git diffs, inspecting security vulnerabilities, code smells, and edge cases.
• /init: Initializes a project-level CLAUDE.md memory file with codebase guidelines and test instructions.
• /bug: Packages active diagnostic logs, reproduction steps, and system info to report tool anomalies.

Parallel Agents and Tooling

• /agents: Lists, inspects, and manages concurrent background subagent workers.
• /mcp: Manages Model Context Protocol (MCP) server connections, inspects available tools, and tests protocol endpoints.
• /help: Displays interactive documentation for all available CLI commands and aliases.
• /logout: Invalidates current credentials and clears stored Anthropic authentication tokens.

Part 5: Core CLI Flags Reference

Launch flags configure execution environments, model tier selection, and automation parameters directly from your shell.

Execution and Model Selection Flags

• claude: Launches Claude Code interactive CLI in the current directory.
• claude -p "<prompt>": Non-interactive batch mode; processes prompt, prints response to stdout, and exits (ideal for CI/CD pipelines).
• claude --model <model-name>: Explicitly overrides default model selection (e.g., claude-3-7-sonnet-20250219).
• claude --dangerously-skip-permissions: Runs agent without prompting for bash execution or file modification approvals (use exclusively in isolated containers).

Session, Memory & Debugging Flags

• claude --continue: Resumes the most recent conversation session in the active repository workspace.
• claude --system-prompt <file>: Appends custom system prompt constraints or architectural compliance guardrails.
• claude --verbose: Enables raw HTTP trace logging, MCP transport messages, and internal agent step telemetry.
• claude --add-dir <path>: Exposes additional directories outside the current git root into the agent's safe read/write perimeter.

Enterprise Best Practices for Agentic CLI Tools

For engineering organizations deploying terminal-based coding agents at scale:

• Define CLAUDE.md Standards: Maintain an authoritative CLAUDE.md in repository roots detailing package build commands, linter conventions, branch naming schemas, and architectural red lines.
• Enforce Permission Boundaries: Avoid using automated permission bypass flags on host workstations containing production credentials. Run unconstrained agents inside Docker sandboxes or ephemeral VM runners.
• Audit Token Economics: Regularly execute /cost and employ prompt caching mechanisms to optimize API token budgets across large development teams.