{"type":"doc","content":[{"type":"heading","attrs":{"id":"921e25cb-9a71-481c-a0c2-83c46829bf06","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"痛点：热榜刷不完，收藏等于没看"}]},{"type":"paragraph","attrs":{"id":"6dd69580-660c-4de1-ba3c-0cbf2b962ea2","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"每天早上想看看 GitHub 上又冒出什么好玩的项目，点开 Trending 页，几十上百个仓库铺下来，star 数、语言、描述一通扫，半小时过去只记住两三个名字，剩下的全进「稍后再看」——然后就再也没有然后了。"}]},{"type":"paragraph","attrs":{"id":"2bcfbc80-ff05-4b02-9b8c-73fc2b2bd905","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"更难受的是，Trending 是按「当日/当周/当月」滚动的，你不看就永远错过了；而且纯列表没有上下文：这个项目解决了什么痛点？和上周那个同类项目比强在哪？值不值得 clone 下来试？这些信息得你自己拼。"}]},{"type":"paragraph","attrs":{"id":"f5cc6d49-facc-44ff-b38a-eedb9f0ab220","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"我后来索性把这件事交给 WorkBuddy：让它每天定时把 GitHub 上最值得关注的新仓库抓出来，用大模型浓缩成一份带「一句话点评 为什么值得看」的早报，推到企业微信，我刷牙的时候扫一眼就完事。"}]},{"type":"heading","attrs":{"id":"5895873c-71a6-4c47-b982-7a2fb7e96231","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"方案架构：抓取 → 聚合 → 沉淀"}]},{"type":"paragraph","attrs":{"id":"43914429-efea-4015-87cc-d845fa104332","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"整个链路不复杂，分三段："}]},{"type":"orderedList","attrs":{"id":"2d3656c5-9b1e-4969-b653-e43f3905816f","start":1,"isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"77539357-6b84-40e2-bf50-7e07bdf42059"},"content":[{"type":"paragraph","attrs":{"id":"4e06b9db-cab1-4984-bc5d-8a9601f63380","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"抓取"},{"type":"text","text":"：用 "},{"type":"text","marks":[{"type":"code"}],"text":"gh api"},{"type":"text","text":" 按时间窗 star 排序捞出近期高增量仓库（GitHub 官方搜索 API，稳定不限裸爬）。"}]}]},{"type":"listItem","attrs":{"id":"b8390cb7-ae12-4bfd-8bc3-df13806a5190"},"content":[{"type":"paragraph","attrs":{"id":"bbd8ba3c-eec2-4c0b-9010-81ea9d9944a2","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"聚合"},{"type":"text","text":"：把仓库名、描述、语言、star 增量丢给大模型，要求它筛掉水文、按「实用度」排序，并给每个项目写一句人话点评。"}]}]},{"type":"listItem","attrs":{"id":"15165dd6-d0d3-489f-8f90-6a9c2f78ea1a"},"content":[{"type":"paragraph","attrs":{"id":"21f29a40-327b-444c-b8ed-f71523c84f5c","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"推送/沉淀"},{"type":"text","text":"：结果落成本地 Markdown，同时推到企业微信机器人；WorkBuddy 的自动化每天早上 8 点自己跑一遍，零人工。"}]}]}]},{"type":"paragraph","attrs":{"id":"a185d17a-4cf5-47d9-9353-20f492776f4e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"WorkBuddy 在这里不是「写个脚本」这么简单，它是编排层：定时触发、调用命令行、把输出喂给模型、再把模型输出格式化推送，全程一个对话里搞定，不用自己拼 CI。"}]},{"type":"heading","attrs":{"id":"6d990871-5c6b-4e37-b6a5-9378bade52a6","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"复制即用的抓取命令"}]},{"type":"paragraph","attrs":{"id":"d5a73d8f-959a-44b7-ab3c-dc8674acf905","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"GitHub Trending 没有官方 API，但"},{"type":"text","marks":[{"type":"bold"}],"text":"搜索 API"},{"type":"text","text":" 能近似还原「近期涨星最快」的效果，而且不会被反爬。直接用 "},{"type":"text","marks":[{"type":"code"}],"text":"gh"},{"type":"text","text":"："}]},{"type":"codeBlock","attrs":{"id":"c492a2ea-3c70-45b6-80cf-71ac49506d90","language":"ja vascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"ja vascript"},"content":[{"type":"text","text":"# 拉取最近 2 天创建、按 star 降序的仓库（约等于 daily trending）ngh api search/repositories n-X GET n-f q='created:>2026-08-05' n-f sort=stars n-f order=desc n-f per_page=30 n--jq '.items[] | "(.full_name)t(.language)t(.stargazers_count)t(.description)"'n"}]},{"type":"paragraph","attrs":{"id":"f76da311-e3be-4700-af30-7a1b6abc7064","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"把日期换成「今天减 2 天」就能每天自动滚。如果你更想要原版 Trending 的那种「跨时段」味道，也可以让 WorkBuddy 用网页抓取 Trending 页（无官方 API，靠 HTML 解析），二者二选一即可。"}]},{"type":"heading","attrs":{"id":"3a647614-c22b-45ce-88cb-933e6ea90f34","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"给大模型的那段 Prompt"}]},{"type":"paragraph","attrs":{"id":"250bc2d5-1533-4cc8-958b-6bb5c05694db","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"把上面命令的输出贴进 WorkBuddy，再附这段指令，它就会吐出一份能直接发的早报："}]},{"type":"codeBlock","attrs":{"id":"e9fc890c-9828-4cbf-86c2-4bee3ed1613b","language":"ja vascript","theme":"atom-one-dark","runtimes":0,"isHoverDragHandle":false,"key":"","languageByAi":"ja vascript"},"content":[{"type":"text","text":"你是一个资深工程师，下面是 GitHub 近期高星增速仓库的裸数据（格式：仓库名语言star数描述）。n请：n1. 过滤掉纯教程/课程/简历模板类仓库；n2. 按「对一线开发者实用度」重新排序，取前 8 个；n3. 每个仓库用一句人话说明：它解决什么痛点、比同类强在哪、适合谁用；n4. 末尾给 3 个「今日最值得 clone」的短名单。n语气克制，不吹不黑，禁营销腔。输出 Markdown 列表即可。n"}]},{"type":"paragraph","attrs":{"id":"76698658-97fb-49d9-8134-077c639cff64","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"实测下来，模型筛水文的准确率挺高，比我自己盲翻 Trending 省心太多。"}]},{"type":"heading","attrs":{"id":"8d76bbb8-1845-4b20-ad08-42f0a8ded8d3","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"踩过的坑"}]},{"type":"bulletList","attrs":{"id":"72828b59-777e-41d0-8912-47afba82ec84","isHoverDragHandle":false},"content":[{"type":"listItem","attrs":{"id":"d2d6afb3-aed8-4acb-a1f9-7833bc0909aa"},"content":[{"type":"paragraph","attrs":{"id":"02f4a3cc-5d41-44ab-b426-d3f3afc1823e","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"裸爬 Trending 会被限流"},{"type":"text","text":"：早期我用网页直抓，频率一高就 429。换成 "},{"type":"text","marks":[{"type":"code"}],"text":"gh api"},{"type":"text","text":" 走 OAuth token 后基本没再被拦，token 用 "},{"type":"text","marks":[{"type":"code"}],"text":"gh auth login"},{"type":"text","text":" 拿就行。"}]}]},{"type":"listItem","attrs":{"id":"1529a684-0eac-40d1-b637-14012abfc233"},"content":[{"type":"paragraph","attrs":{"id":"c55664f6-0873-4a97-8c54-1d53c9cf28c9","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","marks":[{"type":"bold"}],"text":"描述里混用 emoji 和特殊字符：如果直接塞进 Markdown，偶尔会把表格结构搞乱。更稳妥的做法，是让模型输出时先去掉仓库描述里的控制字符。 早报太长，读者基本不会买账：起初一口气发 30 条，后来很快发现，超过 8 条之后，打开率几乎是断崖式下滑。精简到 8 条加 3 个短名单之后，反而每一条都会被认真看完。时区这件事一定别忽视。created:> 里的日期必须按 UTC 来算，不能直接用本地日期，不然凌晨创建的那一批仓库很容易被漏掉。 效果实测 连续跑了一周后发现，每天早上 8:00，企业微信都会准时推来一份结果，平均能筛出 8 个仓库，每条点评大概 30 个字。实际 clone 下来试了 3 个，其中有一个做本地向量检索的小工具，已经顺手接进了日常工作流。真正的收获，其实不只是“多看了几个项目”，更关键的是把“想看却一直没看”的那种心智负担彻底卸掉了——它先帮你看过一遍，剩下的只需要挑自己真正感兴趣的。"42788dfa-f13d-487e-a8ce-a6eb03bc0bab","textAlign":"inherit","indent":0,"level":2,"isHoverDragHandle":false},"content":[{"type":"text","text":"一句话总结"}]},{"type":"paragraph","attrs":{"id":"1055aec9-4211-405a-b76b-134f8e2bd792","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"把 GitHub Trending 的「浏览」动作交给 WorkBuddy 定时抓取 大模型提炼，你只保留「决策」这一步，技术嗅觉就能低成本地每天在线。"}]},{"type":"paragraph","attrs":{"id":"9615a9be-92bb-433f-8422-1b648677c278","textAlign":"inherit","indent":0,"color":null,"background":null,"isHoverDragHandle":false},"content":[{"type":"text","text":"本文是「WorkBuddy 实战」系列之一。"}]}]}","createTime":1786098464,"ext":{"closeTextLink":0,"comment_ban":0,"description":"","focusRead":0},"fa vNum":0,"html":"","isOriginal":0,"likeNum":0,