📥 下载 PDF 电子书 (Download PDF) | 🌐 English Version
💡 AI 原生时代的产品研发与多端编排实战指南
“做产品，最忌讳的是自嗨；用 AI 写代码，最忌讳的是‘人被 AI 牵着走’。本书不讲花哨的学术理论，只聊怎么帮独立开发者和产品经理，用最新的Codex客户端以最快速度搓出能赚钱、能跑通商业闭环的产品。这是‘实战产品说’一贯的硬核实战风格。” —— 主理人 Hunk Wu (X: @ai_pmer)
- Ch.01 告别手写代码：Vibe Coding 时代的产品心智 - 重新思考人机协作边界
- Ch.02 跨端掌控：Codex 多端生产力矩阵搭建 - CLI、桌面端与移动端联调
- Ch.03 破局云端孤岛：沙盒调试与本地环境深度穿透 - 网络、文件挂载与端口映射
- Ch.04 目标驱动：用“边界与断言”驾驭推理型智能体 - 为什么不该教大厨切菜？
- Ch.05 制定 CAP 协议：构建项目专属的 AGENTS.md 规则层 - 制定上下文记忆与代码防腐层
- Ch.06 思维纠偏：如何像技术总监一样透视 CoT 推理链 - 解读思考日志与动态干预
- Ch.07 视觉闭环：Desktop Computer Use 自动巡检与设计还原 - 模拟用户行为，打通视觉端到端测试
- Ch.08 移动看护工作流：全天候离线编排实战 - 随时随地，使用手机微信或飞书监控项目构建
- Ch.09 架构复苏：混乱遗留系统的全景解析与渐进式解耦 - 让 AI 读懂并优化百万行混乱代码
- Ch.10 商业实战：2小时跑通 Next.js + Stripe 商业级 MVP - Next.js 15 + Supabase + Stripe 实战
- Ch.11 触角延伸：Expo 跨端原生 App 开发与云端打包 - 从本地模拟器联调到云端一键打包
- Ch.12 终局思考：独立开发者如何打造自动化商业飞轮 - 一人公司（One-Person SaaS）的流量与变现路径
- Codex 飞书插件：将 Codex 强大的智能体自动化开发与重构能力无缝接入飞书多维表格与机器人工作流，实现日常办公任务的自动化编排与高效数据流转。
开始实践前，建议在项目根目录下配置 AGENTS.md：
# 🤖 Codex Collaboration Protocol (CAP)
## 📌 Project Signature
- Tech Stack: Node.js, React, TypeScript
- Directory Rule: Keep components in `src/components`, logic in `src/hooks`
## 🛑 Hard Constraints
- Never update package.json dependencies without manual approval.
- Do not remove any inline TypeScript documentation or comments.
- Always run `npm run test` before declaring a task complete.本项目持续迭代更新，欢迎在 pmer.cn 官网或公众号“实战产品说”中留言讨论！
本书中提及的部分命令（如 codex --verbose --show-cot、codex refine）、环境变量（如 CODEX_MAX_BUDGET_PER_TASK）以及模型代号（如 GPT-5.5）为针对自主代理（Autonomous Agent）协作流程设计的示意/概念性演示，旨在展示 AI 编排中的“目标驱动”、“思维纠偏”等设计模式与心智模型，并非特定的商业工具或硬性标准。
💡 AI-Native Product Development & Multi-Surface Orchestration
"Product development is never about self-gratification; and in AI coding, the biggest trap is letting the AI lead the human. This book cuts out the fluff. We focus on helping indie hackers and product managers leverage OpenAI Codex to ship monetizable products. This is the hard-boiled style we practice at pmer.cn and '实战产品说'." —— Main Author Hunk Wu (X: @ai_pmer)
- Ch.01 Saying Goodbye to Handwritten Code: Product Mindset in the Era of Vibe Coding
- Ch.02 Cross-Device Control: Building Your Codex Multi-Surface Productivity Matrix
- Ch.03 Breaking the Cloud Island: Sandbox Debugging and Deep Local Environment Tunneling
- Ch.04 Goal-Driven Engineering: Taming Reasoning Agents with Boundaries and Assertions
- Ch.05 Defining the CAP Protocol: Building Your Project's AGENTS.md Rule Compliance Layer
- Ch.06 Correcting Course: Supervising the CoT Reasoning Chain Like a Tech Lead
- Ch.07 Closing the Visual Loop: Automated Auditing and Design Verification with Desktop Computer Use
- Ch.08 Mobile Sentinel Workflows: 24/7 Remote Development and Orchestration
- Ch.09 Codebase Revitalization: Reverse Engineering and Progressive Decoupling of Legacy Systems
- Ch.10 Monetization in Practice: Shipping a Commercial SaaS MVP in 2 Hours
- Ch.11 Mobile Extension: Expo Cross-Platform App Development and Cloud Packaging
- Ch.12 The Final Frontier: Building an Automated Growth Flywheel for a One-Person SaaS
- Codex Feishu Plugin: Integrates Codex's autonomous agent development and refactoring capabilities directly into Feishu (Lark) multidimensional tables and robot workflows, enabling automated daily office task orchestration and high-efficiency data flows.
Before you begin, place AGENTS.md in your project root:
# 🤖 Codex Collaboration Protocol (CAP)
## 📌 Project Signature
- Tech Stack: Node.js, React, TypeScript
- Directory Rule: Keep components in `src/components`, logic in `src/hooks`
## 🛑 Hard Constraints
- Never update package.json dependencies without manual approval.
- Do not remove any inline TypeScript documentation or comments.
- Always run `npm run test` before declaring a task complete.This project is under active development. Join the conversation at pmer.cn or follow Real-World Product Talk (实战产品说) on WeChat!
Some commands (e.g., codex --verbose --show-cot, codex refine), environment variables (e.g., CODEX_MAX_BUDGET_PER_TASK), and model references (e.g., GPT-5.5) mentioned in this book are illustrative/conceptual demonstrations designed to explain autonomous agent orchestration, goal-driven patterns, and thinking chain supervision. They do not represent specific commercial software or mandatory production standards.