Claude Code Cheat Sheet — every command and scenario on one page

The tool that ignited the terminal-agents market when Anthropic released it in February 2025. The idea: open a terminal inside your project, type what you want in plain language, and it reads, edits, runs, and fixes until the job is done.

Install & First Run

Command	What it does
npm install -g @anthropic-ai/claude-code	Install (Node 18+)
brew install claude-code	Install via Homebrew (macOS)
claude	Interactive session in current directory
claude "explain this repo"	Start with a prompt
claude -p "generate a commit message"	Non-interactive: answer and exit
claude -c	Continue last conversation
claude -r	Pick from previous conversations
claude update	Update the CLI
claude --version	Verify version

All Slash Commands

Session Management

Command	Alias	What it does	Scenario
/help	?	All available commands	Forgot a command
/clear	—	Wipe context, fresh start	Between unrelated tasks
/compact	—	Summarize conversation to save tokens	Long job
/model	—	Switch model (Opus / Sonnet / Haiku)	Need stronger/cheaper
/cost	—	Session spend so far	Cost control
/status	—	Current state: model, cost, tokens	Quick check-up
/settings	—	Configuration settings	Change settings
/profile	—	Switch between profiles	Different projects
/exit or Ctrl+D	/quit, q	Quit	Done for the day

Context Management (Most Important!)

Command	What it does	Scenario
/files	Files Claude is currently tracking	See what's in context
/add <file>	Add file to context	Important file not tracked
/drop <file>	Remove file from context	Free up space
/grep <pattern>	Search codebase	Find something
/find <name>	Find file by name	"Where's the auth file?"
/read <file>	Read specific file	Verify file content
/tree	Directory tree	Understand structure

Code Operations

Command	What it does	Scenario
/edit <file>	Open file for editing	Edit this
/diff	Changes Claude made	Review before commit
/undo	Revert last edit	Mistake
/redo	Re-apply reverted edit	Undone by mistake
/review	AI code review	Get AI review
/lint	Run linter	Verify quality
/test	Run tests	Verify nothing broke
/commit	Stage + AI commit	Quick commit

Memory & Configuration

Command	What it does	Scenario
/init	Generate CLAUDE.md	New project
/memory	Edit memory files	Update conventions
/permissions	Manage tool permissions	Control access
/agents	Manage subagents	Delegate work
/hooks	Auto-run commands	Automation
/mcp	MCP server status	Check integrations
/bash	Execute bash directly	Quick command

Plan Mode (For Big Tasks)

Command	Alias	What it does
/plan	/p	Enter Plan Mode
/execute	—	Execute current plan
/next	—	Execute next step
/skip	—	Skip current step
/abort	—	Cancel entire plan
/edit-plan	—	Modify plan
/show	—	Display current plan

Advanced / Hidden

Command	What it does
/editor	Open current task in $EDITOR
/debug	Toggle debug mode
/export	Save conversation as JSON
/tokens	Per-message token breakdown
/vim	Vim editing mode
/delegate "<task>"	Spawn subagent
/agent <id>	Switch to subagent
/kill <id>	Kill subagent
/merge <id>	Merge subagent results

Keyboard Shortcuts

Input Shortcuts

Key	Action
Ctrl+C	Cancel generation mid-run
Ctrl+D	Exit Claude Code
Tab	Autocomplete paths and commands
↑/↓	Navigate command history
Ctrl+R	Search command history
Ctrl+L	Clear screen
Shift+Enter	New line in multi-line input

Approval Shortcuts (when Claude asks)

Key	Action
y or Enter	Yes / Approve once
n	No / Deny
a	Yes to all (remaining session)
d	Show diff before deciding
e	Edit command before executing
q	Quit approval and cancel

Power User Shortcuts

Key	Action
Esc	Interrupt agent (pause, not quit)
Esc twice	Rewind conversation — edit earlier message
Shift+Tab	Cycle permission modes
# at line start	Append note to CLAUDE.md instantly
! at line start	Run bash command, output in chat
@	Mention file — path autocomplete

Approval Modes (Security!)

Full Approval (Default) — Safest

• Every file write: y/n
• Every shell command: y/n
• Every external API: y/n
• Use for: production codebases

Partial Approval — Balanced

• File reads: automatic
• File writes: require approval
• Shell commands: require approval
• Use for: active development

No Approval (--no-approval) — Dangerous ⚠️

• Everything auto-executes
• Use for: isolated containers/test environments ONLY!
• Never on: your main machine

CLI Flags

Flag	Alias	What it does	Scenario
--help	-h	Help
--version	-v	Version
--model <model>	—	Start with specific model	--model opus
--debug	—	Debug mode from start
--no-approval	—	Auto-approve everything ⚠️
--approval-mode <mode>	—	full / partial / none
--profile <name>	—	Start with profile
--plan	—	Start in Plan Mode
--verbose	—	Verbose output
--no-color	—	Disable colors
--context <file>	—	Preload file into context
--execute <command>	—	Run single command and exit	CI/CD
--silent	—	No TUI — output only	scripting
--batch <file>	—	Execute JSON sequence	automation
--offline	—	No network (limited)
--recover	—	Recover session after crash

Configuration

Config Files

File	Purpose
~/.claude/config.json	Global user config
~/.claude/profiles/<name>.json	Profile-specific
./.claude/settings.json	Project-specific
./.claude/ignore	Files to ignore

Example config.json

{
  "model": "claude-3-7-sonnet",
  "approvalMode": "full",
  "autoUpdate": true,
  "theme": "dark",
  "editor": "vim",
  "maxContextTokens": 200000,
  "shell": "/bin/zsh",
  "aliases": {
    "test": "npm test",
    "build": "npm run build"
  }
}

Environment Variables

Variable	Purpose
ANTHROPIC_API_KEY	API key (required)
CLAUDE_MODEL	Default model
CLAUDE_DEBUG	Enable debug
CLAUDE_NO_COLOR	Disable colors
CLAUDE_EDITOR	Default editor
CLAUDE_SHELL	Default shell
CLAUDE_PROFILE	Default profile

Plan Mode — Deep Dive

When to use: Any task > 3 files or > 5 steps

How it works:

1. Describe Goal: You say what you want
2. Claude Proposes Plan: Numbered list of steps
3. Review & Edit: You modify, reorder, or delete
4. Execute: Claude executes step by step with approval
5. Checkpoint: After each step: /undo, /skip, /abort

Example:

> /plan
> Refactor all callback-style async code in src/ to async/await

Claude:
Plan:
1. Identify all files with callback patterns
2. Convert each to async/await
3. Run tests to verify
4. Handle edge cases

> /execute
[Claude executes step 1… approval… step 2… approval…]

Subagents — Branching

Claude Code can spawn subagents — separate instances working on delegated tasks:

Command	What it does
/delegate "<task>" --name <name>	Spawn subagent
/agents	List active subagents
/agent <id>	Switch to subagent
/kill <id>	Kill subagent
/merge <id>	Merge results into main

Patterns:

• Parallel: Subagents on different files simultaneously
• Sequential: Subagent finishes → results fed to parent
• Hierarchical: Parent → subagent → sub-subagent

Hooks — Automation

Git Hooks

claude --install-hook pre-commit

Hook	What it does
pre-commit	Claude reviews staged changes
post-commit	Claude generates PR description
prepare-commit-msg	Claude writes commit message

Custom Hooks

Add to ./.claude/hooks/:

#!/bin/bash
# .claude/hooks/before-edit
npm run lint -- --fix

Event	When it fires
before-edit	Before Claude edits any file
after-edit	After editing
before-command	Before shell command
after-command	After shell command
before-commit	Before commit

MCP (Model Context Protocol)

Built-in MCP Servers

Server	Capability
filesystem	Read/write local files
shell	Execute shell commands
git	Git operations
github	GitHub API
fetch	HTTP requests
browser	Web browsing

Custom MCP Servers

Create ~/.claude/mcp.json:

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": { "DATABASE_URL": "postgresql://..." }
    },
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": { "SLACK_TOKEN": "xoxb-..." }
    }
  }
}

Practical Scenarios

Scenario 1: Refactoring (Plan Mode)

> /plan
> Refactor all callback-style async code in src/ to async/await
[Claude plans… you approve… /execute]

Scenario 2: Debugging

> The tests in auth.test.js are failing. Find and fix the bug.
Claude: I'll run the tests, analyze the error, and trace through...
[Reads → runs → finds bug → fixes → tests]

Scenario 3: Code Review

> Review the changes in this branch for security issues
> /diff
Claude: Analyzing 12 changed files... Found 2 potential SQL injection vectors...

Scenario 4: Documentation

> Generate API documentation from the OpenAPI spec and update the README
[Claude reads spec → generates docs → updates README]

Scenario 5: Learning a Codebase

> Explain the architecture of this project. Start with the entry points.
> @src/main.js what does this do?
> /tree

Scenario 6: Test Generation

> Write comprehensive tests for src/services/payment.js
> Coverage should include edge cases and error handling
[Claude writes tests → runs them → fixes edge cases]

Scenario 7: CI/CD Integration (Non-interactive)

# Non-interactive mode for CI
claude --execute "Review the PR diff and report issues" --no-approval < diff.patch

Scenario 8: Multi-tasking (Subagents)

> /delegate "Audit all API routes for missing auth checks" --name security-audit
> Meanwhile, refactor the database layer to use connection pooling
[Subagent works on security + Claude works on refactor simultaneously]

Scenario 9: Cost Control

> /cost
Session cost: $0.18 · 42.3k tokens

> /compact
[Summarizes conversation → frees context → continues]

Scenario 10: Custom System Prompt

# In ~/.claude/prompts/custom.md:
You are a senior Rust engineer. Always suggest idiomatic solutions.

> /prompt custom
> Rewrite this Go function in Rust

Hidden / Lesser-Known Features

#	Tip	How
1	Silent mode	claude --silent < script.txt — no TUI
2	Batch mode	claude --batch commands.json — command sequence
3	Export conversation	/export > session.json — save for analysis
4	Import context	claude --context previous-session.json
5	Custom prompts	~/.claude/prompts/custom.md + /prompt custom
6	Workspace presets	claude --save-preset "react-app" + --load-preset
7	Token analytics	/tokens — per-message breakdown
8	Diff side-by-side	/diff --side-by-side
9	Auto-save	Saves to ~/.claude/sessions/ — --recover after crash
10	Network isolation	claude --offline — limited functionality
11	Plugin system	claude --plugin @anthropic-ai/plugin-eslint
12	@-mentions	@src/utils/helpers.js or @UserController — focus on code

Troubleshooting

Issue	Solution
"Context full"	/compact or /clear
Slow responses	/drop unused files
Wrong file edits	/undo + be more specific
Claude not finding files	/add to explicitly include
Approval fatigue	Switch to partial approval
API errors	Check ANTHROPIC_API_KEY
Permission denied	Check permissions or use sudo

The Verdict

Top 3 productivity movers (in order):

1. Good CLAUDE.md — write your project conventions, commands, architecture
2. /clear between tasks — clean context = better decisions
3. Plan Mode for big changes — /plan → plans → you approve → executes

Golden Rules:

• Use /compact every 20-30 turns in long sessions
• Use /add to pin critical files in context
• Use @mentions to focus Claude on specific code
• Use Subagents for parallelizable research
• Use Hooks for automated lint/test before commit

Claude Code vs GitHub Copilot CLI:

	Claude Code	Copilot CLI
Concept	"Fix this bug"	"What do I want to do?"
Workflow	Reads → edits → tests → fixes	Generates command → you execute
Multi-file	✅ Deep	❌ One-shot
Plan mode	✅	❌
Subagents	✅	❌
Cost tracking	✅	❌
Git integration	✅ Deep	✅ Surface

Try the live demo below and see the workflow yourself. All the cheat sheets are here, and the full comparison lives on the Models page.