我开发了一个工具,可以将原始的 Git 活动转化为 AI 摘要。

1作者: slmslm5 天前原帖
和许多开发者一样,我厌倦了在不同的代码库之间来回跳转,只为回答一些简单的问题,比如: <p>这周到底发生了什么变化?<p>哪些拉取请求被卡住了?<p>我们发布了什么?<p>谁在等待审核?<p>它位于你的 GitHub/GitLab/Bitbucket 代码库之上,将杂音转化为可读的信息。<p>最困难的部分是什么?Webhooks。 每个提供商发送的负载完全不同: <p>不同的键<p>不同的结构<p>缺失的字段<p>不一致的命名<p>你最终花费更多时间在规范化上,而不是在构建功能上。<p>我们通过创建一个内部统一的事件模式和每个提供商的映射器来解决这个问题。 所有规范化的事件都存储在 MongoDB 中,这非常有帮助,因为灵活的文档模型使得处理略有不同的数据结构变得毫不费力,而不会破坏任何东西。<p>一旦我们有了这个,功能自然就来了: 实时提交和拉取请求监控<p>基于你代码库活动训练的 AI 代理<p>自动化的每周/月总结(通过电子邮件或 Slack)<p>贡献评分的排行榜<p>公开的变更日志页面<p>多平台支持(GitHub/GitLab/Bitbucket)<p>基本上是一个为快速发展的团队提供的统一活动层。<p>为什么要构建它? Git 平台提供原始数据。 团队需要上下文。 开发者想要快速的答案。 管理者想要总结,而不是仪表盘。
查看原文
Like many devs, I got tired of bouncing between repos just to answer simple questions like:<p>What actually changed this week?<p>Which PRs are stuck?<p>What did we ship?<p>Who’s waiting on a review?<p>It sits on top of your GitHub&#x2F;GitLab&#x2F;Bitbucket repos and turns the noise into something readable.<p>The hardest part by far? Webhooks. Each provider sends totally different payloads:<p>different keys<p>different structures<p>missing fields<p>inconsistent naming<p>You end up spending more time normalizing than building features<p>We solved it by creating an internal unified event schema + a mapper for each provider. All normalized events get stored in MongoDB, which helps a lot because the flexible document model makes it painless to handle slightly different shapes of data without breaking anything.<p>Once we had that, the features came naturally Real-time commit &amp; PR monitoring<p>AI agent trained on your repo activity<p>Automated weekly&#x2F;monthly summaries (email or Slack)<p>Leaderboard with contribution scoring<p>Public changelog pages<p>Multi-platform support (GitHub&#x2F;GitLab&#x2F;Bitbucket)<p>Basically a unified activity layer for teams that move fast.<p>Why even build it? Git platforms give you raw data. Teams want context. Devs want fast answers. Managers want summaries, not dashboards.