Picking a stack in a new field means a dozen tabs of back-and-forth — and it's easy to get thrown off by inflated star counts and stale awesome-lists. Clone it, open any AI agent you already use (Claude Code / Codex / Cursor / Cline / Aider), and say "curate the top GitHub projects on X." It searches GitHub broadly, checks each project's real reception, tags it, and writes a localized digest — into a double-click board you can scan in one pass. 进一个新领域选型，你得开十几个标签页来回比，还容易被虚高的 star 和过期的 awesome 清单带偏。Clone 下来，在文件夹里打开你手边任意一个 AI agent（Claude Code / Codex / Cursor / Cline / Aider 都行），说一句「帮我聚合 X 主题的开源项目」——它替你广搜 GitHub、逐个核对真实反响、打标签、写本地化点评，汇成一块双击即开、一眼能扫完的看板。
Star counts alone lie, and awesome-lists keep going stale. GitHub Pick asks one thing of every project — what's its real-world reception right now? — and answers with velocity, recent activity, discussion, and who's integrating it, then lays it out as a board you can actually scan to the end — so you skip the wrong turns and the dead ends.光看 star 会骗人，awesome 清单又总过期。GitHub Pick 对每个项目只问一句——它现在的真实反响到底如何？——拿增速、近期活跃、讨论度、被谁集成这些信号说话，再铺成一块你真能扫读完的看板，帮你少走弯路、不踩老坑。
Afraid of missing a dark horse but drowning in noise? It casts a wide net across GitHub, then scores each candidate on real signals — stars plus velocity, recent activity, newsletter and awesome-list mentions, HN/Reddit/知乎 discussion, who integrates it — each verified with web search. What you get is a checked judgment, not a gut-feel ranking.既怕漏掉黑马、又怕被噪音淹没？它先在 GitHub 上大范围撒网，再用真实信号给每个候选打分——star 加增速、近期活跃、newsletter 和 awesome 清单提及、HN/Reddit/知乎 的讨论、被哪些项目集成——逐条用网页搜索核实。你拿到的是查证过的判断，不是凭感觉的排名。
Worried old projects get cut outright while new ones get buried under star counts? It never pre-filters by creation date: a 2019 veteran and a 2026 upstart stand side by side, and whoever's actually used right now gets in. Age only shows up afterward as a tag, never as an upfront filter — so you don't miss anything still alive and good.怕老项目被一刀切掉、新项目又被 star 数压住？它从不按创建日期预筛：2019 年的老牌和 2026 年的新秀同台，谁现在真有人用谁就进。新旧只在事后用标签标出来，不拿来提前淘汰——任何一个还活着的好东西，你都不会错过。
Don't want to click into every repo to guess if it's still hot or worth learning? Every pick wears 🔥 trending · 🆕 new · 🏆 top-tier · 📜 has-successors — one glance tells you which to pick up and which already has a better replacement, no digging through issues and commit history.不想逐个点进仓库去猜它还火不火、值不值得学？每个精选都挂着 🔥火热 · 🆕新 · 🏆一线 · 📜有继任者，扫一眼就知道哪个该上手、哪个已经有更好的替代，不用再自己翻 issue、看提交记录去判断。
No environment to set up, no server to run — when it's done you double-click index.html and the whole research board is there: projects grouped by category, tagged and ranked. Mark what you've studied, hide what you don't want again; your progress lives in your own browser, so next time you pick up where you left off instead of starting over.不用配环境、不用起服务器，跑完直接双击 index.html 就能看一整块研究看板：项目按分类分组、带标签排好。看过的标「已学」、不想再看的「隐藏」，进度记在你自己浏览器里，下次接着看，不会从头再翻一遍。
Don't want to install one more plugin or learn one more tool for it? It's not a plugin and not a skill you install — just a folder with an instruction file (AGENTS.md): the Claude Code, Codex CLI, Cursor, Cline, or Aider you already use opens it, gets it, and starts. Switch agents later and there's nothing to relearn.不想为了它再装一个插件、再学一套新工具？它不是插件，也不是要安装的 skill，就是一个带说明书（AGENTS.md）的文件夹——你已经在用的 Claude Code、Codex CLI、Cursor、Cline、Aider 打开就能读懂、直接开干。哪天换了 agent，也不用重学。
Don't want to be asked the language every run, or end up with a board mixing English and Chinese? Pick a language on the first run and every digest and the whole board UI stay in it — no re-asking, no toggling back and forth. zh and en work out of the box; any other language gets translated for you the first time.不想每次跑都被追问语言、也不想看板中英文混搭？第一次运行选定语言，之后每一期点评和整块看板 UI 都用它，不再追问、不用来回切。中英文开箱即用，其它语言第一次就给你翻好。
Clone it, open your agent in the folder, and ask:Clone 下来，在文件夹里打开 agent，然后问：
First run, the agent asks which language to lock. When it finishes, open index.html. Prefer manual? Edit config.json and run node scripts/build-data.mjs.第一次运行 agent 会问锁定哪种语言。跑完打开 index.html。想手动？改 config.json 再跑 node scripts/build-data.mjs。