搭建 AI 工作流的标准化方法论——先想清楚再动手。适用于扣子(Coze)、n8n、Dify 等平台。
A standardized methodology for building AI workflows — model first, build second. Works with Coze, n8n, Dify, and similar platforms.
搭工作流最常见的翻车场景:边搭边改,搭到一半发现少了关键节点,回退重来。根本原因是工作流是 DAG——一个节点的输出格式一变,下游全断。
The most common trap: building as you go, discovering a missing node halfway through, ripping it all up. Root cause: workflows are DAGs — change one node's output format and every downstream node breaks.
① 明确最终交付 → ② 确定业务数据 → ③ 排流程节点 → ④ 设人审与异常 → ⑤ 测试、上线与监控
Define the Pin Down the Design Each Set Human Review Test, Ship,
Deliverable Business Data Node & Error Handling and Monitor
| 文件 | 语言 |
|---|---|
| workflow-five-steps.md | 中文 |
| workflow-five-steps.en.md | English |
| README.md | 中英 / CN+EN |
直接阅读对应语言的文件,按五步走即可。
Read the file in your language and follow the five steps.
将 workflow-five-steps.md(或 .en.md)放到你的 Agent Skills 目录。
Place workflow-five-steps.md (or .en.md) in your Agent's Skills directory.
- 扣子(Coze)
- n8n
- Dify
- LangChain / Flowise
- 任何可视化工作流编排工具 / Any visual workflow orchestration tool
- ≤ 5 节点 / nodes:合并第 1+2 步,第 3 步只填功能和依赖 / Merge Steps 1+2; Step 3 — function and dependencies only
- 5–15 节点 / nodes:五步全走 / Run all five steps
- 15+ 节点 / nodes:第 3 步按子流程分组 / Group Step 3 by sub-flow
MIT