Claude Code 2.1 Cheat Sheet

Complete reference for Claude Code CLI tool. Updated for v2.1.x (January 2026).

---

What's New in v2.1

Major Features:

MCP Tool Search — Lazy loading for tools, reducing context usage from ~134k to ~5k tokens
Background Agents — Run up to 15 agents in parallel (Ctrl+B)
Task Management — Built-in task tracking with dependencies (/tasks)
Skill Hot-Reload — Skills update without restart
Forked Context — context: fork for isolated sub-agent execution
Hooks in Frontmatter — Define hooks directly in skills, commands, and agents
Setup Hooks — New event triggered via --init, --init-only, --maintenance
Wildcard Permissions — Bash(npm ), Bash( --help), etc.
PR Review Status — Colored dot showing PR state in prompt footer
Language Setting — Configure Claude's response language
Customizable Keybindings — /keybindings command
Web Session Teleport — /teleport to claude.ai/code
Shift+Enter — Works out-of-the-box (iTerm2, WezTerm, Ghostty, Kitty)

New CLI Flags: --agents, --json-schema, --fallback-model, --max-budget-usd, --remote, --teleport, --init, --maintenance, --chrome, --fork-session, --tools New Slash Commands: /keybindings, /tasks, /plan, /stats, /context, /feedback, /rename, /teleport, /remote-env

---

Core CLI Commands

Command	Description
`claude`	Start interactive REPL
`claude "query"`	One-shot prompt, no REPL
`claude -p "query"`	Print mode — output and exit
`claude -c`	Continue last conversation
`claude --resume <id>`	Resume a specific session
`claude --model <model>`	Use specific model (sonnet/opus/haiku)
`claude --allowedTools <tools>`	Restrict available tools
`claude --agents <path>`	Add custom agent definitions
`claude --init`	Run setup hooks only
`claude --maintenance`	Run maintenance tasks
`claude --teleport`	Send session to claude.ai/code
`claude --fork-session <id>`	Fork from specific session
`claude --remote <url>`	Connect to remote session
`claude mcp`	Manage MCP servers
`claude config`	View/set configuration
`claude config set <key> <value>`	Update config value
`claude config reset`	Reset to defaults

---

CLI Flags (Detailed)

Model & Budget

--model <name> — sonnet (default), opus, haiku
--fallback-model <name> — Model to use if primary fails
--max-budget-usd <amount> — Hard stop at dollar limit
--max-turns <N> — Limit agentic turns (default: 100)

Permissions & Safety

--no-permissions — Use default permissions
--dangerously-skip-permissions — Skip all prompts (use carefully!)
--allowedTools <tool1,tool2> — Whitelist specific tools
--forbiddenTools <tool1,tool2> — Blacklist specific tools

Context & Session

--continue / -c — Continue last conversation
--resume <id> — Resume specific session
--fork-session <id> — Fork from session, keep transcript
--add-dir <path> — Add additional working directory
--remote <url> — Connect to remote session
--teleport — Send to claude.ai/code (web UI)

Output & Verbosity

--print / -p — Print mode (non-interactive)
--verbose — Show full tool calls
--output-format json — JSON output (for scripting)
--json-schema — Output JSON schema for responses

Tools & Agents

--agents <path> — Load custom agent definitions
--tools <plugin:tool> — Load specific MCP tools
--init — Run setup hooks only
--init-only — Setup hooks, then exit
--maintenance — Run maintenance tasks

Browser & Advanced

--chrome — Enable Chrome MCP plugin
--no-cache — Disable prompt caching

---

Slash Commands (Inside REPL)

Core Commands

Command	Description
`/help`	Show all available commands
`/clear`	Clear conversation context
`/compact`	Compress conversation (reduce tokens)
`/cost`	Show token usage and cost
`/stats`	Show detailed session statistics
`/status`	Show session status
`/copy`	Copy last response to clipboard
`/context`	View current context summary

Model & Configuration

Command	Description
`/model`	Switch model mid-conversation
`/permissions`	View and manage permissions
`/keybindings`	Customize keyboard shortcuts
`/rename <name>`	Rename current session
`/feedback`	Send feedback to Anthropic

Memory & Files

Command	Description
`/memory`	Edit CLAUDE.md memory file
`/cd <path>`	Change working directory
`/ls`	List files in current directory

Git & Code Review

Command	Description
`/commit`	Commit staged changes with AI message
`/pr`	Create a pull request
`/review`	Code review current changes
`/review-pr <number>`	Review a PR by number
`/simplify`	Review and simplify changed code

Advanced Features

Command	Description
`/plan`	Enter planning mode (approve before edit)
`/tasks`	View task list and dependencies
`/teleport`	Send session to claude.ai/code
`/remote-env`	Connect to remote environment
`/batch <file>`	Execute batch commands from file

---

Agent Teams (Subagents)

Claude Code can spawn multiple agents to work in parallel or specialize tasks.

Built-in Agent Types

Agent Type	Tools Available	Use Case
`general-purpose`	All tools (Read, Write, Edit, Bash, etc.)	Full implementation tasks
`Explore`	Read-only (Read, Glob, Grep)	Codebase exploration, search
`Plan`	Read-only + planning tools	Architecture planning, design
`statusline-setup`	Read, Edit	Configure status line

Team Workflow

1. Create team — TeamCreate tool 2. Create tasks — TaskCreate with dependencies 3. Spawn teammates — Agent tool with team_name parameter 4. Assign tasks — TaskUpdate with owner field 5. Coordinate — Teammates communicate via SendMessage 6. Shutdown — Send shutdown_request when done

Example: Parallel Work

- Teammate A: Implement backend API
Teammate B: Build frontend UI
Teammate C: Write tests

Task Management Commands

TaskCreate — Create new task
TaskList — List all tasks
TaskUpdate — Update task status/owner
SendMessage — DM or broadcast to teammates

---

Hooks System

Hooks run shell commands on specific events. Configure in .claude/settings.json or skill frontmatter.

Hook Events

Event	When it fires
`PreToolUse`	Before any tool runs
`PostToolUse`	After any tool runs
`Notification`	On agent notifications
`Stop`	When agent stops
`Setup`	On `--init`, `--init-only`, `--maintenance`
`SubmitPrompt`	When user submits prompt

Hook Configuration (settings.json)

{
  "hooks": {
    "PostToolUse": {
      "Write": "echo 'File written: {{file_path}}'",
      "Edit": "prettier --write {{file_path}}"
    },
    "Setup": "npm install && npm run build"
  }
}

Hook Variables

{{file_path}} — File being operated on
{{command}} — Bash command being run
{{content}} — Tool content/result

Frontmatter Hooks (Skills/Commands/Agents)

---
name: my-skill
hooks:
  PreToolUse:
    Bash: echo "Running bash command"
---

---

Keyboard Shortcuts

Default Bindings

Shortcut	Action
`Enter`	Send message
`Shift+Enter`	New line (iTerm2, WezTerm, Ghostty, Kitty)
`Escape`	Cancel current action
`Ctrl+C`	Interrupt generation
`Ctrl+L`	Clear screen
`Ctrl+R`	Search history
`Ctrl+B`	View background agents
`Tab`	Accept autocomplete
`Up/Down`	Navigate history

Customizable Keybindings

Use /keybindings command or edit ~/.claude/keybindings.json:

{
  "toggleTodos": "ctrl+shift+t",
  "toggleTranscript": "ctrl+shift+x",
  "searchHistory": "ctrl+/",
  "chatStash": "ctrl+shift+s"
}

---

Configuration Files

File	Purpose
`CLAUDE.md`	Project instructions (root)
`.claude/settings.json`	Project settings
`~/.claude/CLAUDE.md`	Global user instructions
`~/.claude/settings.json`	Global settings
`~/.claude/keybindings.json`	Custom keyboard shortcuts
`.claude/commands/.md`	Custom slash commands
`.claude/skills//SKILL.md`	Agent skills (auto-invoked)
`.claude/agents/*/AGENT.md`	Custom agent definitions

CLAUDE.md Structure

User preferences and environment context
Project-specific instructions
Code style guidelines
Forbidden actions (e.g., "don't run tests on commit")

settings.json Options

{
  "model": "sonnet",
  "maxTurns": 100,
  "permissionMode": "default",
  "language": "en",
  "hooks": { / ... / },
  "allowedTools": ["Read", "Write", "Edit"],
  "forbiddenTools": ["Bash"]
}

---

Skills vs Commands vs Agents

Commands (User-Invoked)

Location: .claude/commands/<name>.md
Invocation: /command-name
Examples: /commit, /review-pr, /simplify

Frontmatter:

---
allowed-tools: Read, Write, Edit
description: Review and simplify code changes
---

Skills (Model-Invoked)

Location: .claude/skills/<name>/SKILL.md
Invocation: Automatic (model decides)
Examples: claude-developer-platform, frontend-design

Frontmatter:

---
name: frontend-design
description: Build production-grade UIs
triggers:
  - user asks to build web components
  - user mentions React/Vue/Svelte
allowed-tools: Read, Write, Edit, Bash
---

Agents (Custom Subagent Types)

Location: .claude/agents/<name>/AGENT.md
Invocation: Via Agent tool with subagent_type
Examples: Custom test runners, specialized reviewers

Frontmatter:

---
name: test-runner
description: Runs tests and reports results
allowed-tools: Read, Bash
context: fork
---

---

Worktree Patterns

Use EnterWorktree to work in isolated git worktrees:

# User says: "work in a worktree"
claude uses EnterWorktree
Creates: .claude/worktrees/<name>
On exit: prompt to keep or remove

When to Use Worktrees

Experimental changes without affecting main branch
Parallel feature development
Testing risky refactors

---

/simplify Command

Purpose: Review changed code for reuse, quality, and efficiency. Workflow: 1. User invokes /simplify 2. Reviews git diff 3. Identifies: - Duplicate code → Extract helpers - Over-engineering → Simplify abstractions - Missing error handling → Add validation - Performance issues → Optimize Auto-applies fixes — No confirmation needed.

---

/batch Command

Purpose: Execute multiple prompts from a file sequentially. Usage:

# Create batch.txt
echo "Add logging to user.ts" > batch.txt
echo "Write tests for auth.ts" >> batch.txt
echo "Update README with new API" >> batch.txt
Execute
/batch batch.txt

Each line is executed as a separate prompt in order.

---

MCP Plugin System

Model Context Protocol — Extend Claude with external tools.

Manage Plugins

claude mcp add <plugin-name>
claude mcp remove <plugin-name>
claude mcp list

Popular Plugins

filesystem — Enhanced file operations
github — GitHub API integration
postgres — Database queries
brave-search — Web search
chrome — Browser automation (--chrome flag)

Lazy Loading (v2.1)

Tools are loaded on-demand via ToolSearch instead of all-at-once, reducing context from ~134k to ~5k tokens.

---

HTTP API (Remote Sessions)

Run Claude Code as an HTTP server for remote access:

claude --remote http://0.0.0.0:8080

Endpoints:

POST /messages — Send message
GET /sessions — List sessions
GET /sessions/:id — Get session details

Connect from another machine:

claude --remote http://remote-ip:8080

---

Tips & Best Practices

Performance

Use --max-turns to prevent runaway loops
Use /compact to compress context when approaching limits
Prefer Explore agent over general-purpose for read-only tasks

Security

Never use --dangerously-skip-permissions in shared environments
Use .gitignore to exclude .claude/ from version control (or commit selectively)
Review CLAUDE.md before sharing projects

Collaboration

Commit .claude/CLAUDE.md for team-shared instructions
Use hooks to enforce code style (prettier, eslint)
Use /plan mode for high-impact changes

Debugging

Use /stats to see token usage breakdown
Use --verbose to inspect tool calls
Check .claude/sessions/*.jsonl for full transcripts

---

Common Workflows

1. Quick Fix

claude "Fix the bug in auth.ts where tokens expire too soon"

2. Code Review

# Make changes, then:
/review
/simplify
/commit

3. Feature Development

claude
> /plan
> [Review plan]
> Approve
> [Implementation begins]
> /commit
> /pr

4. Multi-Agent Parallel Work

claude
> Create a team to build a REST API with tests
> [Spawns backend, frontend, test agents]
> [Agents work in parallel]
> [Review final output]

---

Troubleshooting

"Context too large" error

Use /compact to compress
Reduce conversation length with /clear
Use --max-turns to limit agent loops

Agent stuck in loop

Ctrl+C to interrupt
Review task list with /tasks
Simplify prompt or break into smaller steps

Hooks not firing

Check .claude/settings.json syntax
Ensure hook scripts are executable
Check hook output in console (--verbose)

MCP plugin not loading

Verify plugin is installed: claude mcp list
Check plugin config in ~/.claude/mcp.json
Try --tools <plugin:tool> to load explicitly

---

Resources

Official Docs: https://docs.claude.ai/code
GitHub: https://github.com/anthropics/claude-code
Changelog: https://github.com/anthropics/claude-code/releases
Community: https://discord.gg/anthropic