Claude Code Cheatsheet

Claude Code Cheatsheet [2026 Updated] – Collabnix

A cheat-sheet is a concise summary of important information meant to be used as a quick reference. In the context of Claude Code, this cheatsheet covers all commonly used keyboard shortcuts, slash commands, CLI flags, MCP server management, agent frontmatter, environment variables, and productivity workflows — so you can move fast without digging through docs.

📋 Table of Contents

⌨️ Keyboard Shortcuts
⚡ Session Slash Commands
🔧 Config Slash Commands
🖥️ CLI Commands & Flags
📁 Memory & Files
🔌 MCP Servers
🤖 Skills & Agents
⚙️ Environment Variables
🔐 Permission Modes
🧠 Workflows & Pro Tips
📣 Community
👉 References

⌨️

Keyboard Shortcuts

General Controls

Shortcut	Description
`Ctrl C`	Cancel current input or generation
`Ctrl D`	Exit the Claude Code session
`Ctrl L`	Clear the screen
`Ctrl O`	Toggle verbose output (see thinking)
`Ctrl R`	Reverse search through history
`Ctrl G`	Open current prompt in external editor
`Ctrl X Ctrl E` NEW	Open in editor (alias for Ctrl G)
`Ctrl B`	Background the running task
`Ctrl T`	Toggle the task list
`Ctrl V`	Paste an image into the prompt
`Ctrl X Ctrl K` NEW	Kill all background agents
`Esc Esc`	Rewind / undo last action

General Controls

Mode Switching & Input

Shortcut	Description
`Shift Tab`	Cycle through permission modes (Normal → Auto-Accept → Plan)
`Alt P`	Switch the active model
`Alt T`	Toggle thinking on / off
`\ Enter`	Insert a newline (quick method)
`Ctrl J`	Insert a newline (control sequence)
`/`	Start a slash command
`!`	Run a direct bash command
`@`	Mention a file with autocomplete

Mode Switching & Input Prefixes

⚡

Session Slash Commands

Command	Description
`/clear`	Clear the current conversation
`/compact [focus]`	Compress context window with optional focus hint
`/resume`	Resume or switch to a previous session
`/rename [name]`	Name the current session
`/branch [name]`	Branch the conversation (alias: /fork)
`/cost`	Show token usage and cost statistics
`/context`	Visualize context usage (grid view)
`/diff`	Open interactive diff viewer
`/copy [N]`	Copy the last (or Nth) response to clipboard
`/rewind`	Rewind conversation / restore code checkpoint
`/export`	Export the full conversation
`/btw <question>`	Ask a side question without consuming context
`/plan [desc]`	Enter plan mode (+ optional auto-start)
`/loop [interval]`	Schedule a recurring task
`/voice`	Enable push-to-talk voice input (20 languages)
`/remote-control` NEW	Bridge session to claude.ai/code (alias: /rc)
`/pr-comments [PR]`	Fetch GitHub PR comments into context
`/security-review`	Run a security analysis of recent changes
`/insights`	Analyze session history and generate a report
`/desktop`	Continue the session in the Desktop app
`/doctor`	Diagnose installation issues
`/feedback`	Submit feedback or bug report (alias: /bug)

Session Slash Commands

🔧

Config Slash Commands

Command	Description
`/config`	Open the settings UI
`/model [model]`	Switch the active model (use ← → to adjust effort)
`/effort [level]` NEW	Set effort level: low / med / high / max / auto
`/fast [on\|off]`	Toggle fast mode
`/vim`	Toggle vim keybindings
`/theme`	Change the color theme
`/permissions`	View and update tool permissions
`/color [color]`	Set the prompt bar color
`/keybindings`	Customize keyboard shortcuts
`/terminal-setup`	Configure terminal keybindings
`/init`	Create a CLAUDE.md file for the project
`/memory`	Edit CLAUDE.md memory files
`/mcp`	Manage MCP servers (interactive UI)
`/hooks`	Manage hooks
`/skills`	List available skills
`/agents`	Manage agents
`/add-dir <path>`	Add an additional working directory
`/reload-plugins`	Hot-reload plugins without restarting
`/usage`	Show plan limits and rate status
`/schedule`	Manage cloud scheduled tasks

Config & Tool Slash Commands

🖥️

CLI Commands & Flags

Core Commands

Command	Description
`claude`	Start an interactive session
`claude "query"`	Start a session with an initial prompt
`claude -p "query"`	Headless / non-interactive mode
`claude -c`	Continue the last conversation
`claude -r "name"`	Resume a named session
`claude update`	Update Claude Code to the latest version
`cat file \| claude -p`	Pipe file content as input (headless)

Core CLI Commands

Key Flags

Flag	Description
`--model`	Set the model to use
`--worktree / -w`	Run in a git worktree
`--name / -n`	Name the session
`--add-dir`	Add an additional working directory
`--agent`	Use a specific agent
`--allowedTools`	Pre-approve a list of tools
`--output-format json`	Return structured JSON output
`--json-schema`	Constrain output to a JSON schema
`--max-turns`	Limit the number of agentic turns
`--max-budget-usd`	Set a cost cap for the session
`--effort`	Set effort: low / med / high / max
`--permission-mode`	Set permission mode at startup
`--verbose`	Enable verbose output
`--bare` NEW	Minimal headless mode (no hooks / LSP)
`--remote`	Open as a web session on claude.ai
`--channels` NEW	Permission relay / MCP push
`--console`	Authenticate via Anthropic Console
`--dangerously-skip-permissions`	⚠️ Skip all permission prompts

Key CLI Flags

📁

Memory & Files

CLAUDE.md Locations

Path	Scope
`./CLAUDE.md`	Project — team-shared, checked into VCS
`~/.claude/CLAUDE.md`	Personal — applies to all projects
`/etc/claude-code/`	Managed — org-wide policy

CLAUDE.md Locations

Rules & Import

Path / Syntax	Description
`.claude/rules/*.md`	Project-level rules (loaded automatically)
`~/.claude/rules/*.md`	User-level rules (loaded globally)
`paths: frontmatter`	Apply rules only to matching file paths
`@path/to/file`	Import another file inside CLAUDE.md
`~/.claude/projects/<proj>/memory/`	Auto-loaded MEMORY.md + topic files

Rules & Auto Memory

💡 Tip: CLAUDE.md survives context compaction — it’s always re-injected after a /compact. Put your most important project context here.

🔌

MCP Servers

Add Servers (–transport)

Transport	Description
`--transport http`	Remote HTTP server (recommended for production)
`--transport stdio`	Local process over stdin/stdout
`--transport sse`	Remote Server-Sent Events stream

MCP Server Transports

Scopes & Config Files

File	Scope
`.claude/settings.local.json`	Local only — not committed to VCS
`.mcp.json`	Project — shared via VCS
`~/.claude.json`	User global — OAuth, MCP state

MCP Configuration Scopes

Management Commands

Command	Description
`/mcp`	Open the interactive MCP management UI
`claude mcp list`	List all registered MCP servers
`claude mcp serve`	Run Claude Code itself as an MCP server

MCP Management

🆕 Elicitation: MCP servers can now request input from the user mid-task using the Elicitation API — great for interactive workflows.

🤖

Skills & Agents

Built-in Skills

Command	Description
`/simplify`	Code review using 3 parallel agents
`/batch`	Large parallel changes with 5–30 git worktrees
`/debug [desc]`	Troubleshoot issues from a debug log
`/loop [interval]`	Schedule a recurring task
`/claude-api`	Load the Anthropic API + SDK reference into context

Built-in Skills

Built-in Agents

Agent	Description
`Explore`	Fast read-only agent (runs on Haiku for speed)
`Plan`	Research-focused agent for plan mode
`General`	Full tool access for complex multi-step tasks
`Bash`	Terminal operations in a separate context

Built-in Agents

Custom Skill Locations

Path	Scope
`.claude/skills/<name>/`	Project-specific skills
`~/.claude/skills/<name>/`	Personal skills (all projects)

Custom Skill Locations

Agent Frontmatter Keys

Key	Description
`permissionMode:`	default / acceptEdits / plan / dontAsk / bypass
`isolation: worktree`	Run the agent in its own git worktree
`memory: user\|project`	Enable persistent memory across sessions
`background: true`	Run as a background task
`maxTurns:`	Limit the number of agentic turns
`model:`	Override the model for this agent
`effort:` NEW	Override effort level for this agent
`initialPrompt:` NEW	Set the opening prompt for the agent
`SendMessage` NEW	Resume agents (replaces the old resume key)

Agent Frontmatter Keys

Skill Frontmatter

Key	Description
`description:`	Trigger phrase for auto-invocation
`allowed-tools:`	Pre-approved tools (skip permission prompts)
`model:`	Override the model for this skill
`effort:` NEW	Override the effort level for this skill
`context: fork`	Run the skill in a subagent context
`$ARGUMENTS`	Placeholder for user-supplied input
`${CLAUDE_SKILL_DIR}`	Reference the skill’s own directory
!`cmd`	Dynamic context injection via shell command

Skill Frontmatter

⚙️

Environment Variables

Variable	Description
`ANTHROPIC_API_KEY`	Your Anthropic API key
`ANTHROPIC_MODEL`	Override the default model
`CLAUDE_CODE_EFFORT_LEVEL`	Default effort: low / med / high
`MAX_THINKING_TOKENS`	Max tokens for thinking; 0 = disabled
`ANTHROPIC_CUSTOM_MODEL_OPTION`	Add a custom entry to the /model picker
`CLAUDE_CODE_MAX_OUTPUT_TOKENS`	Override max output tokens (default 32K)
`CLAUDE_CODE_PLUGIN_SEED_DIR`	Load plugins from multiple directories
`CLAUDECODE`	Set to 1 to signal you are inside a CC shell
`IS_DEMO`	Demo mode — hides email and org info
`CLAUDE_CODE_DISABLE_CRON`	Disable the built-in task scheduler

Key Environment Variables

Config Files

File	Scope
`~/.claude/settings.json`	User settings (global)
`.claude/settings.json`	Project settings (shared via VCS)
`.claude/settings.local.json`	Local overrides (not committed)
`~/.claude.json`	OAuth tokens, MCP state, session data
`.mcp.json`	Project MCP server definitions

Configuration Files

🔐

Permission Modes

Mode	Description
`default`	Prompts for approval before each action
`acceptEdits`	Automatically accepts all file edits
`plan`	Read-only — agent can only plan, not execute
`dontAsk`	Denies actions unless explicitly pre-allowed
`bypassPermissions`	Skips all permission prompts
`--dangerously-skip-permissions`	⚠️ CLI flag to bypass all prompts

Permission Modes

💡 Tip: Use Shift Tab to cycle through Normal → Auto-Accept → Plan modes interactively during a session. Or start a session in a specific mode with --permission-mode plan.

Hook Events

Event	Description
`CwdChanged` NEW	Fires when the working directory changes
`FileChanged` NEW	Fires when a file is modified by the agent

New Hook Events in v2.1.83

🧠

Workflows & Pro Tips

Plan Mode

Action	How
Enter plan mode interactively	Press `Shift Tab` twice
Start session in plan mode	`claude --permission-mode plan`
Max effort for one turn	Type `ultrathink` in your prompt

Plan Mode Tips

Git Worktrees

Command / Key	Description
`claude --worktree name`	Start an isolated branch per feature
`isolation: worktree`	Run an agent in its own worktree (frontmatter)
`sparsePaths` NEW	Checkout only the directories you need
`/batch`	Automatically creates and manages worktrees

Git Worktree Workflow

Context Management

Command / Feature	Description
`/context`	View context usage and optimization tips
`/compact [focus]`	Compress context; focus hint keeps important parts
Auto-compact	Triggers automatically at ~95% context capacity
1M context window	Available on Claude Opus (Max / Team / Enterprise plans)
CLAUDE.md	Always survives compaction — put key context here ✓

Context Management

SDK / Headless

Command	Description
`claude -p "query"`	Non-interactive single query
`--output-format json`	Return structured JSON output
`--max-budget-usd 5`	Hard cost cap for the session
`cat file \| claude -p`	Pipe file content as stdin input
`/loop 5m msg`	Run a recurring task every 5 minutes

SDK & Headless Usage