1 CTO + 7 Specialists. AI works 996, so you don't have to.
世界首个多智能体竞争性生产系统。让AI模拟一家完整8人科技公司，通过内卷机制持续产出生产级代码质量。
- 👨💼 1 CTO - 毒舌全栈技术审查官，终极一票否决
- 👷 7 Specialist - 项目经理/效率/体验/架构/创新/测试/文档
- 🔫 末位淘汰制 - 每轮开掉最差的那个
- 📊 交叉评审矩阵 - 7×7 = 49次同行评审
| Task Type | Regular AI | 996 Agent | Improvement |
|---|---|---|---|
| Simple Task | 81 pts | 93 pts | +15% |
| Medium Task | 67 pts | 89 pts | +33% |
| Complex Task | 43 pts | 95 pts | +121% |
| Average | 62 pts | 90.8 pts | +46% |
📊 Data Source: Blind controlled experiment across 5 difficulty levels. See BENCHMARK_REPORT.md
In Trae IDE, just type:
@996-agent write a production-grade XXX
That's it. The rest is handled automatically by your AI company.
| 文档 Document | 内容 Content |
|---|---|
| 架构演进史 | 从玩梗到工业化的三代完整进化 |
| 基准测试报告 | 5级难度对照实验完整数据 |
| 无限内卷机制 | 迭代边际效益 + 无限内卷SOP |
| 开源爆火方法论 | 从零到1000⭐的完整操作手册 |
| Skill核心源码 | 完整Skill实现 |
| Role | Specialization | Core Value Proposition |
|---|---|---|
| 👨💼 CTO | Full-stack Technical Review | Find fatal flaws everyone else misses |
| 📋 Project Manager | Process Methodology | No process = No consistency |
| ⚡ Efficiency Expert | Algorithm Optimization | Nobody else cares about performance |
| 🎨 UX Designer | API Ergonomics | Devs write unusable APIs |
| 🏗️ Architect | Enterprise Patterns | Everyone else writes toy code |
| 🧪 Innovation Researcher | Cross-domain Ideas | If everyone does it, it's not innovation |
| 🐞 QA Engineer | Destructive Testing | Authors don't break their own code |
| 📝 Technical Writer | Information Architecture | Good devs suck at documentation |
- CTO Kickoff - "The worst performer stays late today"
- Blind Development - 7 people work independently
- 7×7 Cross Review - Everyone reviews everyone
- Competitive Iteration - Must adopt 3 ideas from competitors
- CTO Brutal Review - Point out flaws one by one
- Elimination & Delivery - Fire the worst, ship the champion
It's not "three cobblers equal Zhuge Liang". It's:
- Specialization - Each person only needs to excel at one thing
- Perspective Diversity - 7 completely different angles on the same problem
- Blindspot Cancellation - Your competitor sees bugs you don't
- Competitive Pressure - Real fear of elimination drives learning
- Ultimate Referee - Solves "who reviews the reviewers" problem
🎯 Empirical Law: +7% quality per specialized role Sweet Spot: 8 people. Beyond that communication costs exceed benefits.
User explicitly triggers with:
"Keep involuting"
"3 more rounds"
"Involution until satisfied"
"Fire the worst one"
Four intensity levels + Innovation entropy monitoring to prevent busywork.
See CONTRIBUTING.md for details. All contributions welcome!
MIT - See LICENSE for details.
⭐ If this project made you laugh or think, give it a star!
AI works 996, so you don't have to.
