Claude World Examples

Practical examples and best practices for mastering Claude Code - the official AI-powered CLI from Anthropic

繁體中文

v1.3.0 (2026-10-21): every concept, example, guide, and template is updated for Claude Code 2.1.x, the Claude 5.5 models, and anthropics/claude-code-action@v1, and CI now validates the code samples and links. See the CHANGELOG.

What is Claude World?

Claude World is a community dedicated to helping developers master Claude Code - Anthropic's official CLI for AI-assisted software development. This repository is the companion to claude-world.com: the articles explain the ideas, and the files here are the copy-paste examples.

We focus on:

• Director Mode - A mindset shift from "hands-on developer" to "team director"
• Efficient Workflows - Parallel agents, minimal interruption, autonomous execution
• Best Practices - Battle-tested patterns from real production projects
• Advanced Features - Hooks, MCP, skills, subagents, memory, effort levels, and more

What the Examples Target

Retired model IDs such as claude-sonnet-4-20250514 no longer work; every example uses current IDs.

Repository Structure

claude-world-examples/
├── concepts/           # Core concepts and mental models
├── examples/           # Practical code examples
├── guides/             # In-depth guides and tutorials
├── templates/          # Ready-to-use templates (CLAUDE.md, workflows)
└── case-studies/       # Real-world project case studies

Most pages have a Traditional Chinese translation in a zh-TW/ folder next to them.

Content Overview

Core Concepts

Practical Examples

Guides

Templates

Case Studies

Topics Covered on claude-world.com

The website covers these advanced topics in depth:

Quick Start

1. Install Claude Code

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Homebrew
brew install --cask claude-code

On Windows, run irm https://claude.ai/install.ps1 | iex in PowerShell or winget install Anthropic.ClaudeCode. Installing with npm still works but is deprecated. Update later with claude update.

2. Create a CLAUDE.md in your project

Run /init inside Claude Code to generate one from your codebase, or start from this:

# Project Instructions

## Tech Stack
- [Your framework]
- [Your database]

## Development Guidelines
- Run tests before committing
- Use TypeScript strict mode

## Autonomous Operations
- Read any file
- Run tests
- Create/modify source files

## Requires Confirmation
- Push to remote
- Delete files
- Modify environment variables

CLAUDE.md is guidance. To enforce the "Requires Confirmation" list, add permission rules to .claude/settings.json:

{
  "$schema": "https://json.schemastore.org/claude-code-settings.json",
  "permissions": {
    "ask": ["Bash(git push *)", "Bash(rm *)"],
    "deny": ["Read(./.env)", "Read(./.env.*)"]
  }
}

3. Start Claude Code

claude

The Director Mode Mindset

Traditional approach (Hands-on Developer Mode):

You: "Can you help me fix this bug?"
Claude: "Sure, I can help. First, let me ask..."
You: [waits for questions]
Claude: [asks 5 questions]
You: [answers questions]
Claude: [finally starts working]

Director Mode approach:

You: "Fix the login bug"
Claude: [immediately launches 3 parallel agents]
       [Agent 1: searches for auth code]
       [Agent 2: checks error logs]
       [Agent 3: reviews recent changes]
       [finds and fixes the bug]
       [commits with clear message]
You: [reviews the completed work]

Key principle: You're the Director. Claude is your autonomous team. Define goals, let them execute.

The Three Pillars

1. Goals over Instructions - Define outcomes, not steps
2. Team Delegation - Trust parallel agents to work concurrently
3. Outcome Verification - Review results, not every action

Model Selection

Choose the right model for each task:

Quick tips: Switch with /model (or Option+P on macOS, Alt+P on Windows/Linux) and set reasoning depth with /effort. In the Claude API, use the full IDs claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5, and claude-fable-5-1.

Quality Checks

Every pull request runs the Validate workflow:

• yamllint and actionlint (with shellcheck) on every workflow template
• Code blocks: every json code block must be valid JSON, every yaml block must parse, and complete workflows inside Markdown are linted with actionlint
• lychee checks every link in every Markdown file (also weekly, to catch link rot)

See CONTRIBUTING.md to run the same checks locally.

Community

• Website: claude-world.com
• Discord: Claude Code Builders Taiwan
• Twitter: @lukashanren1

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

How to Contribute

1. Examples: Add practical examples showing patterns in action
2. Templates: Create CLAUDE.md templates for different frameworks
3. Guides: Write in-depth guides on specific topics
4. Translations: Help translate content to other languages

For discussions and questions, join our Discord community.

Security

Found a problem in an example or template that could harm someone who copies it? Please report it privately; see SECURITY.md.

License

MIT License - see LICENSE

Built with Claude Code by the Claude World community.

Last updated: 2026-10-21