Agent Architecture Learning Skill

English | 中文

---

中文版本

简介

这是一个教学型 AI Skill，帮助产品架构师通过框架对比真正内化 Agent 架构决策逻辑。

"不教写代码，教做决策"

背景

在学习 Agent 开发过程中，我发现通过对比不同框架来学习是最有效的方法。但市面上大多数教程都在教"怎么写代码"，而不是"怎么做架构决策"。

这个 Skill 专门为产品架构师设计，专注于：

• 判断产品需要什么级别的 Agent 能力
• 理解框架设计的核心权衡
• 识别 AI 给出的技术方案是否过度工程

功能特性

• 自动初始化：分析文章/框架的定位
• 诊断对话：快速定位知识盲区
• 引导学习：用业务场景解释概念
• 记忆机制：增量学习，持续追踪掌握状况
• 总结沉淀：每次学习后生成记录
• 跨平台兼容：同时支持 OpenAI 和 Claude

核心框架：四阶模型

第一阶：大脑（LLM） → 能思考，不能行动
第二阶：手脚（工具） → 能行动，做完就忘
第三阶：经验（记忆） → 能记住，能自主探索
第四阶：协作（控制） → 用户能实时干预

核心判断：你的产品需要演进到第几阶？

安装

# 复制 skill 文件到 skills 目录
cp agent-architecture-learning.md ~/.claude/skills/

# （可选）创建学习历史文件
cp agent-learning-history.md ~/.claude/knowledge/

使用

在对话中直接调用：

/skill agent-architecture-learning

或提供任何 Agent 框架相关的文章/文档链接。

交付物

agent-architecture-learning-skill/
├── README.md                          # 本文件
├── agent-architecture-learning.md     # 主 skill 文件
└── docs/
    ├── learning-history-template.md   # 学习记录模板
    ├── framework-comparison.md        # 框架对比参考
    └── scenario-library.md            # 业务场景库

技术栈

• AI 平台：OpenAI (ChatGPT) + Anthropic (Claude)
• 格式：YAML front matter + Markdown
• 存储：本地文件系统

作者

Murphy (@bor799)

• 产品架构师，追求极致效率
• GitHub: https://github.com/bor799

License

MIT

---

English Version

Introduction

A teaching-oriented AI Skill that helps product architects truly internalize Agent architecture decision logic through framework comparison.

"Don't teach coding, teach decision-making"

Background

While learning Agent development, I discovered that learning by comparing different frameworks is the most effective method. However, most tutorials focus on "how to write code" rather than "how to make architecture decisions."

This Skill is designed specifically for product architects, focusing on:

• Determining what level of Agent capability a product needs
• Understanding core trade-offs in framework design
• Identifying whether AI-generated technical solutions are over-engineered

Features

• Auto-initialization: Analyze article/framework positioning
• Diagnostic dialogue: Quickly identify knowledge gaps
• Guided learning: Explain concepts using business scenarios
• Memory mechanism: Incremental learning with continuous progress tracking
• Summary & retention: Generate records after each session
• Cross-platform compatibility: Support for both OpenAI and Claude

Core Framework: Four-Stage Model

Stage 1: Brain (LLM) → Can think, cannot act
Stage 2: Hands (Tools) → Can act, forgets after completion
Stage 3: Experience (Memory) → Can remember, can explore autonomously
Stage 4: Collaboration (Control) → User can intervene in real-time

Core Question: What stage does your product need to evolve to?

Installation

# Copy skill file to skills directory
cp agent-architecture-learning.md ~/.claude/skills/

# (Optional) Create learning history file
cp agent-learning-history.md ~/.claude/knowledge/

Usage

Invoke directly in conversation:

/skill agent-architecture-learning

Or provide any Agent framework-related article/documentation link.

Deliverables

agent-architecture-learning-skill/
├── README.md                          # This file
├── agent-architecture-learning.md     # Main skill file
└── docs/
    ├── learning-history-template.md   # Learning record template
    ├── framework-comparison.md        # Framework comparison guide
    └── scenario-library.md            # Business scenario library

Tech Stack

• AI Platforms: OpenAI (ChatGPT) + Anthropic (Claude)
• Format: YAML front matter + Markdown
• Storage: Local file system

Author

Murphy (@bor799)

• Product Architect, pursuing extreme efficiency
• GitHub: https://github.com/bor799

License

MIT