🚀 This methodology is now a real product! CodeFlow Product Page · GitHub · Try PWA
Just tell the PM what you need, go grab a coffee, and come back to review the results.
你只需要跟 PM 说清楚要做什么，然后去喝杯咖啡，回来验收成果。
Build a 4-role AI team (PM + DEV + OPS + QA) in Cursor IDE. The AIs collaborate autonomously — developing, deploying, and testing on their own. You only talk to the PM.
在 Cursor IDE 中搭建 PM + DEV + OPS + QA 四角色 AI 团队，AI 之间自主协同——自动开发、自动部署、自动测试，人类只需和 PM 沟通任务。
Battle-tested: 87 person-days of work in 17 days, 91 production deployments, zero incidents.
实战验证：17 天完成 87 人天工作量，线上发版 91 次，零事故。
| Content | |
|---|---|
| Ch.1 | Why split into roles — limits of single-agent AI |
| Ch.2 | Step-by-step setup: directory structure → role definitions → patrol rules → 4 chat tabs |
| Ch.3 | Core innovation: Filename as Protocol — zero databases, zero message queues, 7 fields in one filename |
| Ch.4 | Task flow: 7-step closed loop from assignment → dev → deploy → test → archive |
| Ch.5 | Work standards: each role's "rules of engagement" + mandatory documentation |
| Ch.6 | Auto patrol bot: screen image recognition + event-driven, with full Python source (280 lines) |
| Ch.7 | Real-world results: production data + 9 running screenshots |
No database, no message queue, no API — one filename carries all routing information:
TASK-20260329-006-PM01-to-DEV01.md
│ │ │ │ │ │ │
│ │ │ │ │ │ └── Format: Markdown
│ │ │ │ │ └────── Recipient: DEV-01
│ │ │ │ └────────── Direction: PM → DEV
│ │ │ └──────────────── Sender: PM-01
│ │ └──────────────────── Sequence: 6th task of the day
│ └───────────────────────────── Date: 2026-03-29
└────────────────────────────────── Type: Task ticket
7 fields, 0 database tables, 0 lines of config code.
You: "Do a round of security hardening."
PM-01: "Got it. Breaking it down now."
— Go do something else —
PM-01 breaks down tasks → writes tickets to tasks/
DEV-01 auto picks up → codes, self-tests, submits report
PM-01 auto reviews → creates deploy task
OPS-01 auto deploys → health check, writes report
PM-01 auto assigns → creates test task
QA-01 auto tests → security + stress tests, writes report
PM-01 auto archives → all done
You come back: "Done?"
PM-01: "All complete. Here's the report."
PM-01 reviewing code changes item by item
DEV-01 auto-creates Todo list and codes
Patrol bot auto-monitoring + event-driven notifications
tasks/ directory — filenames in action
├── cursorAI-automated-team-EN.md # Full tutorial — English (960+ lines)
├── cursorAI-automated-team-CN.md # Full tutorial — Chinese (1000+ lines)
├── auto_patrol.py # Patrol bot source code (280 lines)
├── README.md # This file
├── roles/ # 📋 Role definition files (ready to use)
│ ├── PM-01.md # PM 项目经理 + 架构师（中文）
│ ├── PM-01-EN.md # PM: Project Manager + Architect (English)
│ ├── PM-01-Work-Standards-CN.md # PM 工作规范（中文）
│ ├── PM-01-Work-Standards.md # PM Work Standards (English)
│ ├── DEV-01.md # DEV 全栈开发工程师（中文）
│ ├── DEV-01-EN.md # DEV: Full-stack Developer (English)
│ ├── OPS-01.md # OPS 运维部署工程师（中文）
│ ├── OPS-01-EN.md # OPS: Operations Engineer (English)
│ ├── QA-01.md # QA 质量测试工程师（中文）
│ └── QA-01-EN.md # QA: Quality Assurance Engineer (English)
└── images/ # Screenshots
- Using Cursor for development and want more AI automation
- Managing multiple AI Agents that need to collaborate autonomously
- Need complete task tracking and audit trails
- Solo developers / small teams using AI to replace repetitive work
- IDE: Cursor (Agent mode)
- Patrol bot: Python 3.10 + pyautogui + pyperclip
- Communication: File system + Markdown (zero external dependencies)
- AI models: Claude / GPT (choose per role)
Full UI automation patrol bot included: auto_patrol.py — 280 lines, ready to run.
© 2026 joinwell52-AI. All rights reserved.
- Non-commercial use: You may copy, distribute, and modify this content for non-commercial purposes (personal learning, teaching, public sharing), provided that:
- Original author attribution and source links are retained;
- Content is not distorted, tampered with, or used for illegal purposes.
- Commercial use: Strictly prohibited, including but not limited to:
- Using content in paid courses, paid documents, commercial websites/accounts;
- Integrating content into commercial products, services, or marketing materials;
- Directly or indirectly obtaining commercial benefit from this content.
- Violation of these terms without written authorization from the author may result in legal action.
Author: joinwell52-AI
From real production project experience. 2026-03-29