awesome-agentic-ai-zh-learning

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

awesome-agentic-ai-zh Learning Skill

awesome-agentic-ai-zh 学习技能指南

Skill by ara.so — AI Agent Skills collection.

由 ara.so 提供的技能指南 — AI Agent 技能合集。

Overview

概述

awesome-agentic-ai-zh

is a comprehensive, structured learning roadmap for AI Agent development that takes you from LLM basics to building multi-agent systems. It provides:

Two learning tracks: Track A (CLI Power User) and Track B (Agent Builder)
8 core stages with 145+ curated projects and resources
27 hands-on exercises with working code examples
Bilingual content (Traditional Chinese, Simplified Chinese, English)
5 specialized branches for researchers, developers, teachers, knowledge workers, and everyday users

The project is particularly valuable for understanding the Claude Code ecosystem (MCP, Skills, Plugins, Subagents) and modern agent interfaces (Computer Use, Browser Use, Code Sandbox).

awesome-agentic-ai-zh

是一份全面、结构化的AI Agent开发学习路线图，涵盖从LLM基础到构建多智能体系统的完整路径。它提供：

两条学习路径：路径A（CLI高级用户）和路径B（智能体构建者）
8个核心阶段，包含145+精选项目与资源
27个实操练习，附可运行代码示例
双语内容（繁体中文、简体中文、英文）
5个专业分支，面向研究者、开发者、教师、知识工作者及普通用户

该项目在理解Claude Code生态系统（MCP、Skills、Plugins、Subagents）和现代智能体接口（Computer Use、Browser Use、Code Sandbox）方面极具价值。

Installation & Setup

安装与配置

bash

undefined

bash

undefined

Clone the repository

git clone https://github.com/WenyuChiou/awesome-agentic-ai-zh.git cd awesome-agentic-ai-zh

No additional dependencies for reading the roadmap

Individual exercises may require Python and specific libraries


For complete setup (first-time learners):
```bash


针对零基础学习者的完整配置步骤：
```bash

Read the setup guide first

先阅读配置指南

cat resources/setup-guide.md

Install Python 3.8+ if needed

若需要，安装Python 3.8+

python --version

Install common dependencies for exercises

安装练习所需的通用依赖

pip install anthropic openai langchain chromadb

undefined

pip install anthropic openai langchain chromadb

undefined

Learning Path Structure

学习路径结构

Shared Foundation (Stage 0-2)

共享基础阶段（第0-2阶段）

Stage 0: Foundations (

stages/00-foundations.md

)

Python, CLI, git, API basics, JSON
Duration: 1-2 weeks

Stage 1: LLM Basics (

stages/01-llm-basics.md

)

Token concepts, API usage, LLM comparison, local LLM (Ollama)
Duration: 1 week

Stage 2: Prompt Engineering (

stages/02-prompt-engineering.md

)

System prompts, few-shot learning, Chain-of-Thought
Duration: 1-2 weeks

第0阶段：基础能力 (

stages/00-foundations.md

)

Python、CLI、git、API基础、JSON
时长：1-2周

第1阶段：LLM基础 (

stages/01-llm-basics.md

)

Token概念、API使用、LLM对比、本地LLM（Ollama）
时长：1周

第2阶段：提示词工程 (

stages/02-prompt-engineering.md

)

系统提示词、少样本学习、思维链（Chain-of-Thought）
时长：1-2周

Track A: CLI Power User

路径A：CLI高级用户

bash

undefined

bash

undefined

Navigate Track A

浏览路径A内容

cat tracks/cli/A1-cli-intro.md # CLI agent comparison & setup cat tracks/cli/A2-cli-workflow.md # Workflow patterns cat tracks/cli/A3-cli-production.md # Production integration

cat tracks/cli/A1-cli-intro.md # CLI智能体对比与配置 cat tracks/cli/A2-cli-workflow.md # 工作流模式 cat tracks/cli/A3-cli-production.md # 生产环境集成

Key resource

核心资源

cat resources/cli-agents-guide.md


**Total duration**: 8-10 weeks (including shared foundation)

cat resources/cli-agents-guide.md


**总时长**：8-10周（含共享基础阶段）

Track B: Agent Builder

路径B：智能体构建者

bash

undefined

bash

undefined

Navigate Track B

浏览路径B内容

cat stages/03-tool-use-and-hello-agent.md # Function calling, ReAct cat stages/04-agent-frameworks.md # LangGraph, AutoGen, CrewAI cat stages/05-claude-code-ecosystem.md # MCP, Skills, Plugins (SHARED HUB) cat stages/06-memory-rag.md # Context engineering, RAG cat stages/07-multi-agent-production.md # Multi-agent orchestration cat stages/07.5-advanced-agentic-concepts.md # Advanced concepts (reading) cat stages/08-agent-interfaces.md # Computer Use, Browser Use (SHARED HUB)


**Total duration**: 16-22 weeks minimum, 5-7 months realistically (5-8 hrs/week)

cat stages/03-tool-use-and-hello-agent.md # 工具调用、ReAct cat stages/04-agent-frameworks.md # LangGraph、AutoGen、CrewAI cat stages/05-claude-code-ecosystem.md # MCP、Skills、Plugins（共享中心） cat stages/06-memory-rag.md # 上下文工程、RAG cat stages/07-multi-agent-production.md # 多智能体编排 cat stages/07.5-advanced-agentic-concepts.md # 进阶概念（阅读） cat stages/08-agent-interfaces.md # Computer Use、Browser Use（共享中心）


**总时长**：至少16-22周，实际需5-7个月（每周5-8小时）

Key Commands & Navigation

关键命令与导航

Finding Resources

查找资源

bash

undefined

bash

undefined

List all stage files

列出所有阶段文件

ls stages/

View glossary of terms

查看术语表

cat resources/glossary.md

Check CLI agents comparison

查看CLI智能体对比

cat resources/cli-agents-guide.md

Browse exercises

浏览练习内容

ls exercises/stage-*/

undefined

ls exercises/stage-*/

undefined

Running Exercises

运行练习

Each stage has 1-5 exercises in

exercises/stage-X/

python

undefined

每个阶段在

exercises/stage-X/

目录下有1-5个练习：

python

undefined

Example: Stage 1 - First LLM API call

示例：第1阶段 - 首次LLM API调用

exercises/stage-1/01-first-llm-call/main.py

import os from anthropic import Anthropic

def main(): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain AI agents in one sentence."}
    ]
)

print(message.content[0].text)

if name == "main": main()


```bash

import os from anthropic import Anthropic

def main(): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain AI agents in one sentence."}
    ]
)

print(message.content[0].text)

if name == "main": main()


```bash

Set up environment

设置环境变量

export ANTHROPIC_API_KEY="your-key-here"

Run exercise

运行练习

python exercises/stage-1/01-first-llm-call/main.py

undefined

python exercises/stage-1/01-first-llm-call/main.py

undefined

Dual-Path SDK Examples

双路径SDK示例

Most exercises provide both Anthropic SDK and Ollama (local) implementations:

python

undefined

大多数练习同时提供Anthropic SDK和Ollama（本地）实现：

python

undefined

Using Anthropic Claude

使用Anthropic Claude

from anthropic import Anthropic

client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY")) response = client.messages.create( model="claude-3-5-sonnet-20241022", max_tokens=1024, messages=[{"role": "user", "content": "Hello"}] )

from anthropic import Anthropic

Using Ollama (local)

使用Ollama（本地）

import ollama

response = ollama.chat( model="llama3.2", messages=[{"role": "user", "content": "Hello"}] )

undefined

import ollama

response = ollama.chat( model="llama3.2", messages=[{"role": "user", "content": "Hello"}] )

undefined

Configuration Patterns

配置模式

API Key Management

API密钥管理

bash

undefined

bash

undefined

Set environment variables (recommended)

设置环境变量（推荐方式）

export ANTHROPIC_API_KEY="sk-ant-..." export OPENAI_API_KEY="sk-..."

Or use .env file

或使用.env文件

cat > .env << EOF ANTHROPIC_API_KEY=sk-ant-... OPENAI_API_KEY=sk-... EOF

Load in Python

在Python中加载

from dotenv import load_dotenv load_dotenv()

undefined

from dotenv import load_dotenv load_dotenv()

undefined

Local LLM Setup (Ollama)

本地LLM配置（Ollama）

bash

undefined

bash

undefined

Install Ollama

安装Ollama

curl -fsSL https://ollama.com/install.sh | sh

Pull a model

拉取模型

ollama pull llama3.2

Test

测试

ollama run llama3.2 "Explain what an AI agent is"

undefined

ollama run llama3.2 "Explain what an AI agent is"

undefined

Common Usage Patterns

常见使用模式

Pattern 1: Following the Learning Path

模式1：跟随学习路径

bash

undefined

bash

undefined

For complete beginners

完全零基础用户

cat stages/00-foundations.md

→ Complete exercises in exercises/stage-0/

→ 完成exercises/stage-0/中的练习

Then proceed sequentially

然后按顺序推进

cat stages/01-llm-basics.md cat stages/02-prompt-engineering.md

Choose your track

选择你的路径

cat tracks/cli/A1-cli-intro.md # OR cat stages/03-tool-use-and-hello-agent.md

undefined

cat tracks/cli/A1-cli-intro.md # 或者 cat stages/03-tool-use-and-hello-agent.md

undefined

Pattern 2: Quick Reference for Specific Topics

模式2：特定主题快速参考

bash

undefined

bash

undefined

Need MCP information?

需要MCP相关信息？

cat stages/05-claude-code-ecosystem.md

Need multi-agent patterns?

需要多智能体模式？

cat stages/07-multi-agent-production.md

Need Computer Use examples?

需要Computer Use示例？

cat stages/08-agent-interfaces.md

undefined

cat stages/08-agent-interfaces.md

undefined

Pattern 3: Building Your First Agent

模式3：构建你的第一个智能体

Follow the comprehensive walkthrough:

bash

cat walkthroughs/build-first-agent-in-7-steps.md

Example from the walkthrough (Stage 3: Tool Use):

python

undefined

遵循完整的分步指南：

bash

cat walkthroughs/build-first-agent-in-7-steps.md

指南中的示例（第3阶段：工具使用）：

python

undefined

exercises/stage-3/02-function-calling/main.py

import os import json from anthropic import Anthropic

def get_weather(city: str) -> dict: """Mock weather API - returns fake data""" return { "city": city, "temperature": 22, "condition": "sunny" }

def main(): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

tools = [{
    "name": "get_weather",
    "description": "Get current weather for a city",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "City name"}
        },
        "required": ["city"]
    }
}]

response = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    tools=tools,
    messages=[
        {"role": "user", "content": "What's the weather in Tokyo?"}
    ]
)

# Handle tool use
if response.stop_reason == "tool_use":
    tool_use = next(block for block in response.content if block.type == "tool_use")
    
    if tool_use.name == "get_weather":
        result = get_weather(**tool_use.input)
        
        # Send result back
        response = client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=1024,
            tools=tools,
            messages=[
                {"role": "user", "content": "What's the weather in Tokyo?"},
                {"role": "assistant", "content": response.content},
                {
                    "role": "user",
                    "content": [{
                        "type": "tool_result",
                        "tool_use_id": tool_use.id,
                        "content": json.dumps(result)
                    }]
                }
            ]
        )

print(response.content[0].text)

if name == "main": main()

undefined

import os import json from anthropic import Anthropic

def get_weather(city: str) -> dict: """模拟天气API - 返回模拟数据""" return { "city": city, "temperature": 22, "condition": "sunny" }

def main(): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

tools = [{
    "name": "get_weather",
    "description": "获取城市当前天气",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "城市名称"}
        },
        "required": ["city"]
    }
}]

response = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    tools=tools,
    messages=[
        {"role": "user", "content": "What's the weather in Tokyo?"}
    ]
)

# 处理工具调用
if response.stop_reason == "tool_use":
    tool_use = next(block for block in response.content if block.type == "tool_use")
    
    if tool_use.name == "get_weather":
        result = get_weather(**tool_use.input)
        
        # 返回结果
        response = client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=1024,
            tools=tools,
            messages=[
                {"role": "user", "content": "What's the weather in Tokyo?"},
                {"role": "assistant", "content": response.content},
                {
                    "role": "user",
                    "content": [{
                        "type": "tool_result",
                        "tool_use_id": tool_use.id,
                        "content": json.dumps(result)
                    }]
                }
            ]
        )

print(response.content[0].text)

if name == "main": main()

undefined

Pattern 4: ReAct Agent Implementation

模式4：ReAct智能体实现

python

undefined

python

undefined

exercises/stage-3/04-react-agent/main.py

import os from anthropic import Anthropic

def search_papers(query: str) -> list: """Mock paper search""" return [ {"title": "Attention Is All You Need", "year": 2017}, {"title": "BERT: Pre-training of Deep Bidirectional Transformers", "year": 2018} ]

def summarize_paper(title: str) -> str: """Mock paper summarizer""" return f"Summary of '{title}': A foundational paper in NLP..."

def react_loop(user_query: str, max_iterations: int = 5): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

tools = [
    {
        "name": "search_papers",
        "description": "Search academic papers by query",
        "input_schema": {
            "type": "object",
            "properties": {
                "query": {"type": "string"}
            },
            "required": ["query"]
        }
    },
    {
        "name": "summarize_paper",
        "description": "Get summary of a paper by title",
        "input_schema": {
            "type": "object",
            "properties": {
                "title": {"type": "string"}
            },
            "required": ["title"]
        }
    }
]

messages = [{"role": "user", "content": user_query}]

for i in range(max_iterations):
    response = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=2048,
        tools=tools,
        messages=messages
    )
    
    messages.append({"role": "assistant", "content": response.content})
    
    if response.stop_reason == "end_turn":
        return response.content[0].text
    
    # Execute tools
    tool_results = []
    for block in response.content:
        if block.type == "tool_use":
            if block.name == "search_papers":
                result = search_papers(**block.input)
            elif block.name == "summarize_paper":
                result = summarize_paper(**block.input)
            
            tool_results.append({
                "type": "tool_result",
                "tool_use_id": block.id,
                "content": str(result)
            })
    
    messages.append({"role": "user", "content": tool_results})

return "Max iterations reached"

import os from anthropic import Anthropic

def search_papers(query: str) -> list: """模拟论文搜索""" return [ {"title": "Attention Is All You Need", "year": 2017}, {"title": "BERT: Pre-training of Deep Bidirectional Transformers", "year": 2018} ]

def summarize_paper(title: str) -> str: """模拟论文摘要生成""" return f"Summary of '{title}': A foundational paper in NLP..."

def react_loop(user_query: str, max_iterations: int = 5): client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))

tools = [
    {
        "name": "search_papers",
        "description": "Search academic papers by query",
        "input_schema": {
            "type": "object",
            "properties": {
                "query": {"type": "string"}
            },
            "required": ["query"]
        }
    },
    {
        "name": "summarize_paper",
        "description": "Get summary of a paper by title",
        "input_schema": {
            "type": "object",
            "properties": {
                "title": {"type": "string"}
            },
            "required": ["title"]
        }
    }
]

messages = [{"role": "user", "content": user_query}]

for i in range(max_iterations):
    response = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=2048,
        tools=tools,
        messages=messages
    )
    
    messages.append({"role": "assistant", "content": response.content})
    
    if response.stop_reason == "end_turn":
        return response.content[0].text
    
    # 执行工具
    tool_results = []
    for block in response.content:
        if block.type == "tool_use":
            if block.name == "search_papers":
                result = search_papers(**block.input)
            elif block.name == "summarize_paper":
                result = summarize_paper(**block.input)
            
            tool_results.append({
                "type": "tool_result",
                "tool_use_id": block.id,
                "content": str(result)
            })
    
    messages.append({"role": "user", "content": tool_results})

return "Max iterations reached"

Usage

使用示例

result = react_loop("Find papers about transformers and summarize the most important one") print(result)

undefined

result = react_loop("Find papers about transformers and summarize the most important one") print(result)

undefined

MCP (Model Context Protocol) Integration

MCP（Model Context Protocol）集成

Stage 5 covers the Claude Code ecosystem. Key MCP concepts:

python

undefined

第5阶段涵盖Claude Code生态系统。核心MCP概念示例：

python

undefined

Example MCP server structure

MCP服务器结构示例

See stages/05-claude-code-ecosystem.md for details

详情见stages/05-claude-code-ecosystem.md

from mcp.server import Server from mcp.types import Tool, TextContent

app = Server("my-mcp-server")

@app.tool() async def get_document(doc_id: str) -> str: """Fetch document by ID""" # Your implementation return f"Document content for {doc_id}"

@app.tool() async def search_database(query: str) -> list: """Search internal database""" # Your implementation return [{"id": "1", "title": "Result"}]

undefined

from mcp.server import Server from mcp.types import Tool, TextContent

app = Server("my-mcp-server")

@app.tool() async def get_document(doc_id: str) -> str: """通过ID获取文档""" # 你的实现代码 return f"Document content for {doc_id}"

@app.tool() async def search_database(query: str) -> list: """搜索内部数据库""" # 你的实现代码 return [{"id": "1", "title": "Result"}]

undefined

Using MCP with Claude Desktop

在Claude Desktop中使用MCP

json

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "my-server": {
      "command": "python",
      "args": ["/path/to/your/mcp_server.py"]
    }
  }
}

json

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "my-server": {
      "command": "python",
      "args": ["/path/to/your/mcp_server.py"]
    }
  }
}

Specialized Branches

专业分支

For Researchers

面向研究者

bash

cat branches/for-researcher.md

Focus: Literature review automation, paper writing assistance, multi-agent review systems

bash

cat branches/for-researcher.md

重点：文献综述自动化、论文写作辅助、多智能体评审系统

For Developers

面向开发者

bash

cat branches/for-developer.md

Focus: Cursor/Aider integration, CLI delegation, code review agents

bash

cat branches/for-developer.md

重点：Cursor/Aider集成、CLI委托、代码评审智能体

For Everyday Users

面向普通用户

bash

cat branches/for-everyday-users.md

Focus: Using ChatGPT/Claude.ai effectively, privacy scenarios, CLI agent introduction (no coding required)

bash

cat branches/for-everyday-users.md

重点：高效使用ChatGPT/Claude.ai、隐私场景、CLI智能体入门（无需编码）

Troubleshooting

故障排除

Common Issues

常见问题

Issue: "API key not found"

bash

undefined

问题："API key not found"

bash

undefined

Verify environment variable

验证环境变量

echo $ANTHROPIC_API_KEY

Set it if missing

若缺失则设置

export ANTHROPIC_API_KEY="sk-ant-..."


**Issue: "Module not found"**
```bash

export ANTHROPIC_API_KEY="sk-ant-..."


**问题："Module not found"**
```bash

Install required package

安装所需包

pip install anthropic

For exercises requiring multiple packages

针对需要多个包的练习

pip install -r requirements.txt


**Issue: "Ollama connection refused"**
```bash

pip install -r requirements.txt


**问题："Ollama connection refused"**
```bash

Check if Ollama is running

检查Ollama是否运行

curl http://localhost:11434/api/tags

Start Ollama if needed

若未运行则启动

ollama serve


**Issue: "Which stage should I start from?"**
- Have Python/git basics? → Start Stage 1
- Complete beginner? → Start Stage 0
- Want to use CLI agents without coding? → Go directly to Track A (A1-cli-intro.md)

ollama serve


**问题："我应该从哪个阶段开始？"**
- 具备Python/git基础？→ 从第1阶段开始
- 完全零基础？→ 从第0阶段开始
- 想使用CLI智能体但不想编码？→ 直接进入路径A（A1-cli-intro.md）

Stage-Specific Help

阶段专属帮助

bash

undefined

bash

undefined

Each stage has a glossary

每个阶段都有术语表

cat resources/glossary.md

Setup troubleshooting

配置故障排除

cat resources/setup-guide.md

CLI-specific issues

CLI相关问题

cat resources/cli-agents-guide.md

undefined

cat resources/cli-agents-guide.md

undefined

Best Practices

最佳实践

Follow the path sequentially — Each stage builds on previous knowledge
Complete the exercises — 27 hands-on exercises are designed for learning by doing
Use dual-path approach — Try both cloud APIs (Anthropic/OpenAI) and local models (Ollama)
Check the glossary —
```
resources/glossary.md
```
has all terminology in Chinese + English
Join the community — The project welcomes contributions and questions
Set realistic expectations — Track B takes 5-7 months part-time; Track A takes 8-10 weeks

按顺序学习 — 每个阶段的知识都基于前一阶段
完成所有练习 — 27个实操练习专为边做边学设计
采用双路径方式 — 同时尝试云API（Anthropic/OpenAI）和本地模型（Ollama）
查阅术语表 —
```
resources/glossary.md
```
包含所有中英双语术语
加入社区 — 项目欢迎贡献和提问
设置合理预期 — 路径B需业余时间学习5-7个月；路径A需8-10周

Example: Complete Agent Build

示例：完整智能体构建

See

walkthroughs/build-first-agent-in-7-steps.md

for a 350-line Paper Summary Bot that evolves from Stage 1 to Stage 7, demonstrating:

LLM API basics (Stage 1)
Prompt engineering (Stage 2)
Tool use and ReAct (Stage 3)
Framework integration with LangGraph (Stage 4)
Memory and RAG (Stage 6)
Multi-agent orchestration (Stage 7)

查看

walkthroughs/build-first-agent-in-7-steps.md

，了解一个从第1阶段到第7阶段逐步演进的350行论文摘要机器人，展示：

LLM API基础（第1阶段）
提示词工程（第2阶段）
工具使用与ReAct（第3阶段）
LangGraph框架集成（第4阶段）
记忆与RAG（第6阶段）
多智能体编排（第7阶段）

Additional Resources

额外资源

bash

undefined

bash

undefined

Full glossary

完整术语表

cat resources/glossary.md

Diagrams

图表资源

ls resources/diagrams/

- learning-map.png

- branch-decision-tree.png

- banner.png

Complete setup guide

完整配置指南

cat resources/setup-guide.md

undefined

cat resources/setup-guide.md

undefined

Project Metadata

项目元数据

License: MIT
Languages: Traditional Chinese (primary), Simplified Chinese, English
Scope: 145+ curated projects, 27 exercises, 8 stages, 2 tracks, 5 branches
Repository: https://github.com/WenyuChiou/awesome-agentic-ai-zh
Stars: 1462+ (as of 2026-05-16)

许可证：MIT
语言：繁体中文（主要）、简体中文、英文
规模：145+精选项目、27个练习、8个阶段、2条路径、5个分支
仓库地址：https://github.com/WenyuChiou/awesome-agentic-ai-zh
星标数：1462+（截至2026-05-16）