全 AI 工作流程
2 分•作者: fagnerbrack•6 个月前
我刚刚构建了一个系统，在 pre-push 阶段，它使用 CLI 监控对 GH Actions 的更改，并将错误反馈给 Claude。我在 5 个不同的编辑器中克隆了代码库，并通过提示让它始终开发、提交和推送可运行的代码。它严重依赖静态分析和使用 Biome 进行代码检查，以确保最大的正确性和严格的测试规范。
代码库是一个单体仓库，其中包含可以独立开发的内聚服务。采用基于主干的开发模式，更改直接推送到生产环境（有一个部署后检查会运行冒烟测试）。
我不确定是否有人构建过类似的东西，但基本上，我可以在不同的文件夹中 "git clone" 代码库的新版本，并且只需花几分钟创建一个提示，就能字面意义上地复制自己。
99% 的工作是设计提示并调整各个部分，使其相互独立，这样我就可以适应上下文窗口的最新部分。
我靠，这是什么鬼？！
查看原文
So I just built a system that on pre-push, watch changes to GH actions using CLI and feed errors to Claude. I am cloning the repo in 5 separate editors with a prompt that makes it develop, commit and push working code at all times. It leans heavily on static analysis and linting with Biome to ensure maximum correctness and strong guidelines for testing.<p>Repository is a monorepo which has cohesive services that can be developed independently. Trunk-Based development and changes go straight to production (there's a post-deploy check that runs smoke tests)<p>I'm not sure if anyone else has built something like this, but essentially I can "git clone" a new version of the repository in a different folder and literally multiply myself by just spending a couple minutes creating a prompt<p>99% of the Job is designing prompts and adjusting the pieces to be independent so I can fit in the recent part of the context window<p>WHAT THE ACTUAL FUCK?