Claude Code Cheat Sheet: Commands, Shortcuts, and Quick Reference

Claude Code now accounts for roughly 4% of all GitHub commits, according to 2026 usage tracking data. With that level of adoption, the gap between a developer who knows /compact and one who doesn't is the gap between hitting context limits mid-task and finishing cleanly.

The official Anthropic cheatsheet lists commands alphabetically, which works for lookup but not for building habits. This guide reorganizes everything by workflow: session management, context control, code operations, and external tools. Whether you run Claude Code for solo sessions or Claude Cowork collaboration across agents, knowing the command surface turns exploratory fumbling into focused work.

To see every command available in your current session, type / on an empty prompt. The autocomplete menu shows built-in defaults, custom skills, plugin commands, and tools from connected MCP servers. Your list will vary depending on what you've installed.

Session and Navigation - /help lists all available commands

• /status shows your account, active model, working directory, and version
• /clear wipes conversation history and starts fresh (project memory stays)
• /resume reopens a previous session to continue where you left off
• /login and /logout handle authentication and account switching
• /exit quits the CLI

Context Management

Context is the most common bottleneck in long sessions. These commands keep you under the limit:

• /compact summarizes the conversation to free up context window space
• /context shows what's currently loaded in the context window
• /cost displays token usage and spend for the session
• /usage shows plan limits and current rate-limit status
• /btw asks a quick side question without adding to the main thread

Code and File Operations

• /init explores your codebase and generates a starter CLAUDE.md
• /diff opens an interactive viewer of uncommitted changes and per-turn diffs
• /plan enters Plan Mode, optionally with a task description
• /simplify runs parallel review agents to check recent changes for quality issues
• /add-dir grants file access to an additional directory for the session
• /rewind rolls the conversation and code back to an earlier checkpoint

Configuration - /model shows or switches the active model (Opus, Sonnet, Haiku, Fable)

• /config opens settings for theme, defaults, and editor mode
• /permissions views or changes which tools need approval before running
• /memory opens CLAUDE.md files in scope for editing
• /hooks shows hook configuration for tool events

External Tools and Agents

• /mcp manages MCP server connections and authentication
• /agents lists, creates, or edits subagents
• /skills lists available skills in the session

Output and Diagnostics

• /copy copies the last response or a code block to clipboard
• /export saves the conversation to a file or clipboard
• /feedback reports an issue to Anthropic with session context
• /doctor diagnoses install and environment problems

These work in the Claude Code CLI. Terminal support varies, so press ? in any session to see what's available in your setup.

Interrupting and navigating:

• Esc interrupts Claude mid-response so you can type again
• Esc, Esc (double-tap) opens the rewind and checkpoint menu
• Ctrl + C cancels the current input, or exits on an empty prompt
• Ctrl + R reverse-searches through your prompt history
• Up / Down arrows scroll through previous prompts

Display and modes:

• Ctrl + O expands to the verbose, full transcript view
• Shift + Tab cycles permission mode: default, acceptEdits, plan
• ? shows shortcuts available in your terminal or IDE

Input helpers:

• @ followed by a path references a file or directory in your prompt
• / on an empty line opens the command menu with autocomplete

Shift + Tab deserves special attention. Instead of typing /plan to enter plan mode, one keystroke cycles through all three permission modes. In acceptEdits mode, Claude auto-approves file writes without prompting. In plan mode, Claude explores and proposes but won't execute. Cycling back returns to default, where each action requires approval.

Give Your Claude Code Agent Persistent Storage

Fast.io workspaces keep agent-generated files indexed and searchable across sessions. Connect through MCP, build your output, and hand ownership to a human when the work is done. Start with a 14-day free trial.

Start 14-Day Trial

Claude Code's three permission modes control how much autonomy the agent has:

Default mode asks for approval on file edits and tool calls. This is what you get when you launch claude without flags.

Accept Edits mode auto-approves file writes for the rest of the session. Good for refactoring sessions where you trust the direction but don't want to approve every save. Switch with Shift + Tab or configure as a default in /config.

Plan mode puts Claude in read-only exploration. It reads files, searches code, and proposes changes, but executes nothing. Start here when diagnosing a bug or exploring unfamiliar code. Enter with /plan or cycle to it with Shift + Tab.

Effort Levels

The /effort command controls how much reasoning Claude applies per response:

• low keeps responses short and fast for simple lookups
• medium balances speed and depth (the default)
• high adds deeper analysis for complex questions
• xhigh and max enable extended thinking for multi-file architectural problems
• auto lets Claude pick based on the prompt complexity

Higher effort uses more tokens but produces better results for debugging and multi-step refactoring. For quick "where is this function defined?" questions, low is enough.

CLAUDE.md and Project Memory

CLAUDE.md is a markdown file Claude reads at the start of every session. Place one in your project root with coding conventions, file structure notes, or patterns to avoid. Use /memory to edit it, or /init to generate a starter file from your codebase.

Claude Code also maintains auto-memory across conversations, saved to ~/.claude/projects/<project>/memory/. Corrections you make in one session carry forward automatically, which is especially useful in Claude Cowork sessions where multiple agents share context about the same codebase.

Hooks

Hooks are shell commands that fire at specific points in Claude's lifecycle. Define them in settings.json:

• SessionStart loads context when a session launches
• PreToolUse runs before tool execution (good for logging bash commands)
• PostToolUse runs after (useful for auto-formatting saved files)
• Stop runs checks when Claude finishes a task
• PostCompact re-injects critical instructions after context compression

Skills and Subagents

Skills are packaged instructions for specific workflows. Store them at .claude/skills/<name>/SKILL.md and invoke as slash commands. Custom agents live in .claude/agents/ and launch with claude --agent=<name>.

Subagents handle focused subtasks in their own context. For parallel work, set isolation: worktree in agent frontmatter to give each subagent its own git worktree.

The /mcp command manages connections to external services through the Model Context Protocol. MCP servers extend what Claude Code can do beyond file editing and bash:

claude mcp add <server-name>

Common servers include GitHub (issues and pull requests), Sentry (error tracking), and database connectors. Each server exposes tools that Claude calls during a session.

For workflows that produce artifacts like research, generated documentation, or data exports, files can end up scattered across local directories with no easy way to share them. Fast.io workspaces solve this through an MCP server at mcp.fast.io that gives Claude Code agents persistent, searchable storage. Files uploaded during one session stay indexed and available in the next. An agent can build a workspace, organize output, and transfer ownership to a human when the work is done.

To connect:

claude mcp add fastio /storage-for-agents/

The MCP approach keeps your agent's capabilities modular. Add servers as you need them, remove what you don't. Type /mcp to see active connections and manage authentication.