[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-github-skills-repos-turn-ai-coding-into-workflows-zh":3,"article-related-github-skills-repos-turn-ai-coding-into-workflows-zh":35,"series-tools-4ecd2dc5-dc13-4f9b-86bd-0a97f2f814d7":87},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":18,"translated_content":10,"views":19,"is_premium":20,"created_at":21,"updated_at":21,"cover_image":11,"published_at":22,"rewrite_status":23,"rewrite_error":10,"rewritten_from_id":24,"slug":25,"category":26,"related_article_id":27,"status":28,"google_indexed_at":10,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":29,"topic_cluster_id":33,"embedding":34,"is_canonical_seed":20},"4ecd2dc5-dc13-4f9b-86bd-0a97f2f814d7","GitHub 技能庫把 AI 寫程式變流程","\u003Cp data-speakable=\"summary\">我把 \u003Ca href=\"\u002Ftag\u002Fgithub\">GitHub\u003C\u002Fa> 趨勢報告拆成一套可直接複製的 AI 寫程式流程，重點是把\u003Ca href=\"\u002Fnews\u002Fibm-prompt-guide-turns-ai-guesses-into-outputs-zh\">提示\u003C\u002Fa>詞變成技能與規則。\u003C\u002Fp>\u003Cp>我用 AI 寫程式工具一陣子了，老實說，最煩的不是它不會寫，是它太會亂寫。你叫它先補測試，它說好；你叫它別亂動 production 相關檔案，它也說好；你叫它照團隊規範走，它還是會在你沒注意時自己發明一套。前幾分鐘看起來很像會做事，十分鐘後就開始像一個拿到 root 權限的實習生，還很有自信。這次我看到 GitHub 趨勢報告後，才比較確定問題在哪裡：不是模型不夠強，是我們一直用聊天方式在管流程，當然會失控。\u003C\u002Fp>\u003Cp>這份洞察的起點是 DEV \u003Ca href=\"\u002Fnews\u002Fmicrosoft-copilot-2026-update-real-workflows-zh\">Co\u003C\u002Fa>mmunity 上的 \u003Ca href=\"https:\u002F\u002Fdev.to\u002Fyanceyxin\u002Fgithub-weekly-trending-repositories-report-2g68\">GitHub Weekly Trending Repositories Report\u003C\u002Fa>，資料來源還串了 \u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F\">star-history.com\u003C\u002Fa>。我不是要把它吹成什麼神作，但它很直接地點出一件事：大家正在把 \u003Ca href=\"\u002Ftag\u002Fai-coding\">AI coding\u003C\u002Fa> 從「會講話」推向「會照規則做事」。\u003C\u002Fp>\u003Ch2>我看到的不是熱度，是大家終於受不了亂跑的代理人\u003C\u002Fh2>\u003Cblockquote>AI agent skills frameworks have taken over the open-source ecosystem.\u003C\u002Fblockquote>\u003Cp>翻譯一下就是：開源社群現在不太想再跟 AI 玩即興表演了。大家要的是重複得出來的行為，不是每次都靠運氣。這句話我很有感，因為我自己最常踩的坑，就是同一個工具在不同專案裡表現完全兩樣。今天在 A repo 很乖，明天到 B repo 就開始亂補架構、亂改檔名、亂猜需求。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779134707718-sslr.png\" alt=\"GitHub 技能庫把 AI 寫程式變流程\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>我之前真的有一段時間以為，只要 p\u003Ca href=\"\u002Fnews\u002Ftop-ai-prompt-engineering-tools-2026-zh\">romp\u003C\u002Fa>t 寫得夠完整，模型就會自動變穩。結果不是。只要對話一長、上下文一多，它就開始忘記前面講過的限制。你越想靠一句話把事情講清楚，它越容易在後面自己發揮。\u003C\u002Fp>\u003Cp>實操寫法很簡單：把你每週都會重複講的規則，從聊天框搬到檔案裡。不要再每次都臨時打字。改成 repo 裡有一份固定的規則文件，讓 AI 每次都先讀。這樣做看起來很土，但真的比你每次重講一次有效。\u003C\u002Fp>\u003Cul>\u003Cli>把重複出現的要求寫成檔案，不要只放在對話裡。\u003C\u002Fli>\u003Cli>每個規則只管一件事，別一份文件塞十種人格。\u003C\u002Fli>\u003Cli>先管住危險行為，再談效率。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>mattpocock\u002Fskills 會紅，不是因為會吹，是因為夠像真實工作\u003C\u002Fh2>\u003Cp>報告裡最醒目的專案是 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmattpocock\u002Fskills\">mattpocock\u002Fskills\u003C\u002Fa>。作者把它描述成「來自 .\u003Ca href=\"\u002Ftag\u002Fclaude\">claude\u003C\u002Fa> 目錄的技能」。這句話我覺得很誠實，因為它不是在賣概念，是直接把平常真的會用到的規則攤開來。\u003C\u002Fp>\u003Cp>這裡的重點不是「技能」這個詞聽起來很潮，而是它把 AI 的行為拆成可重用的模組。像是測試優先、除錯步驟、危險指令停下來確認、特定語言的慣例，這些都不是靈感，是工作規則。你如果還在用一大段大雜燴 prompt 去管所有情境，效果通常都不太行。\u003C\u002Fp>\u003Cp>我自己以前也很愛寫長 prompt，覺得自己像在做提示工程。後來才發現，那比較像在寫一封很累的信。真正有用的是把規則做成小塊，讓工具可以每次都照著走。你不用期待它變聰明，你只要讓它不要一直犯同一種蠢就好。\u003C\u002Fp>\u003Cp>實操寫法：\u003C\u002Fp>\u003Cul>\u003Cli>把常用任務拆成小技能，例如測試、除錯、重構、安全操作。\u003C\u002Fli>\u003Cli>每個技能都要有「好輸出」跟「壞輸出」範例。\u003C\u002Fli>\u003Cli>技能名稱要像命令，不要像論文標題。\u003C\u002Fli>\u003Cli>如果團隊每週都在抱怨同一件事，那就是你該寫技能的地方。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Claude Code 周邊爆出來，代表大家在補工具洞\u003C\u002Fh2>\u003Cp>報告裡不只一個 repo 在講 \u003Ca href=\"\u002Ftag\u002Fclaude-code\">Claude Code\u003C\u002Fa>，還有像 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Ffarion1231\u002Fcc-switch\">cc-switch\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faffaan-m\u002Feverything-claude-code\">everything-claude-code\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Faddyosmani\u002Fagent-skills\">addyosmani\u002Fagent-skills\u003C\u002Fa>，甚至 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fskills\">anthropics\u002Fskills\u003C\u002Fa> 也在做同一件事。這不是巧合，這是生態開始長出周邊了。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779134714102-ncsq.png\" alt=\"GitHub 技能庫把 AI 寫程式變流程\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>白話講，工具只要真的進入日常工作，就不會只剩主產品本身。你一定會開始需要環境切換、規則管理、範本、教學、整合方式。這些東西看起來不性感，但它們才是讓工具能不能落地的差別。\u003C\u002Fp>\u003Cp>我自己最有感的是，很多 \u003Ca href=\"\u002Ftag\u002Fai-工具\">AI 工具\u003C\u002Fa> demo 的時候都超順，一到真實專案就卡。不是因為模型不行，是因為你還沒把周邊補齊。專案規範、權限、測試流程、部署風險，這些東西才是日常。你不把它們包進去，AI 就只會在最理想的狀況下表現正常。\u003C\u002Fp>\u003Cp>實操寫法：\u003C\u002Fp>\u003Cul>\u003Cli>先補環境切換與專案初始化流程。\u003C\u002Fli>\u003Cli>把常見規則集中成一個入口，不要散在 Slack、Notion、腦袋裡。\u003C\u002Fli>\u003Cli>讓團隊每個人用同一套起手式，別各自發明版本。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>spec-kit 其實是在反對「憑感覺寫」\u003C\u002Fh2>\u003Cp>報告裡另一個我很在意的是 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgithub\u002Fspec-kit\">github\u002Fspec-kit\u003C\u002Fa>。它走的是 spec-driven development，也就是先寫規格，再排計畫，再拆任務，最後才進實作。這條路線我很認同，因為它直接把很多 AI 寫 code 的毛病卡掉了。\u003C\u002Fp>\u003Cp>翻譯一下就是：不要一開始就叫模型生 code。先叫它把需求講清楚，再叫它拆步驟，最後才動手。這樣做不是慢，是少走冤枉路。你如果直接讓它寫，八成會得到一坨看起來像樣、實際上很難 review 的東西。\u003C\u002Fp>\u003Cp>我之前遇過一個很典型的狀況：我只想改一個 API 行為，結果 AI 幫我順手重構三個檔案，還很開心地說這樣更乾淨。問題是我根本沒要求它這樣做。這就是沒有規格的後果，工具會自己補完故事，而且通常補得很爛。\u003C\u002Fp>\u003Cp>實操寫法很直接：\u003C\u002Fp>\u003Cul>\u003Cli>先要求 AI 產出需求摘要。\u003C\u002Fli>\u003Cli>再要它列限制條件與假設。\u003C\u002Fli>\u003Cli>接著拆成可 review 的任務清單。\u003C\u002Fli>\u003Cli>最後才允許它寫程式。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Hermes Agent 提醒我：記憶比嘴砲重要\u003C\u002Fh2>\u003Cp>報告也提到 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\">NousResearch\u002Fhermes-agent\u003C\u002Fa>，它主打會跟著你一起成長，還提到自我改善記憶、Python 架構、MIT 授權。這種方向我其實不意外，因為大家早晚都會碰到同一個問題：AI 每次都從零開始，真的很煩。\u003C\u002Fp>\u003Cp>如果一個代理人沒有記憶，它就只能在每次對話裡重跑一次新手教學。這對問答還行，對軟體工作就很差。因為軟體工作有很多固定偏好、團隊規範、歷史決策，這些都不該每次重講。你講一次、兩次、三次，最後還是會懷疑是不是自己在跟失憶症工具共事。\u003C\u002Fp>\u003Cp>我現在比較相信的做法不是等模型自己學會，而是把記憶外掛在專案裡。決策紀錄、除錯筆記、團隊規則、風險提醒，這些都應該留在 repo，不要只留在某個人的腦袋。這樣就算工具換了，流程還在。\u003C\u002Fp>\u003Cp>實操寫法：\u003C\u002Fp>\u003Cul>\u003Cli>建立決策紀錄檔，記下為什麼這樣做。\u003C\u002Fli>\u003Cli>把常見錯誤與修正方式寫成可查的筆記。\u003C\u002Fli>\u003Cli>如果工具支援持久記憶，就只存專案層級資訊。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>技能庫其實是文件，只是更能被執行\u003C\u002Fh2>\u003Cp>我覺得這波最值得注意的地方，不是某個 repo 多紅，而是大家開始把知識做成可執行文件。像 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmultica-ai\u002Fandrej-karpathy-skills\">multica-ai\u002Fandrej-karpathy-skills\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Ffinancial-services\">anthropics\u002Ffinancial-services\u003C\u002Fa> 這類例子，說穿了就是把領域知識包成 AI 看得懂的規則。\u003C\u002Fp>\u003Cp>這件事的價值很務實。以前很多團隊的流程都藏在老鳥腦中，新人只能靠問。現在如果你把規則寫成技能，AI 可以先幫你守第一層，新人也能照著做，不會每次都從零摸索。這不是偷懶，是把隱性知識變成明文規則。\u003C\u002Fp>\u003Cp>我自己的經驗是，只要團隊沒寫下來，AI 幫忙的品質就會非常飄。每個人餵它的方式不同，結果就會長得很不一致。你以為是模型不穩，其實是輸入來源根本沒統一。\u003C\u002Fp>\u003Cp>實操寫法：\u003C\u002Fp>\u003Cul>\u003Cli>把常見流程寫成 living docs，不要寫完就放生。\u003C\u002Fli>\u003Cli>每次出現新失敗模式，就補進技能文件。\u003C\u002Fli>\u003Cli>有合規、資安、部署限制的領域，直接寫進規則，不要靠口頭提醒。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>可抄的模板\u003C\u002Fh2>\u003Cpre>\u003Ccode># AI Agent 技能包模板（繁中版）\n\n這份模板可以直接放進你的 repo，給 Claude Code、Cursor、Codex 或任何會讀專案檔案的 AI 助手用。\n\n## 1) agent-instructions.md\n\n# Agent 操作規則\n\n在這個 repo 裡工作時，請先讀取對應技能檔案，再開始產出內容。\n\n## 基本規則\n- 先確認需求，再開始寫 code。\n- 先寫測試，再做實作。\n- 有危險指令時先停下來確認。\n- 變更要小，方便 review。\n- 如果需求不清楚，只問一個最關鍵的問題。\n- 如果任務太大，先拆成步驟。\n\n## 2) repo-skills\u002Ftdd.md\n\n# 技能：測試優先\n\n當你要新增或修改行為時：\n1. 先找出最小的失敗測試。\n2. 先寫測試。\n3. 再做最小實作讓測試通過。\n4. 測試通過後才考慮重構。\n5. 除非使用者明講，否則不要跳過測試步驟。\n\n## 3) repo-skills\u002Fdebugging.md\n\n# 技能：結構化除錯\n\n當東西壞掉時：\n1. 先重述症狀。\n2. 列出最可能的原因。\n3. 先查成本最低的原因。\n4. 一次只改一個變數。\n5. 記錄哪些原因已經排除。\n\n你要回報：\n- 哪裡壞了\n- 你檢查了什麼\n- 你改了什麼\n- 為什麼這個修正合理\n\n## 4) repo-skills\u002Fsafety.md\n\n# 技能：安全與防呆\n\n在執行危險指令前，先停下來請求確認。\n\n危險指令範例：\n- git push --force\n- rm -rf\n- 生產環境資料庫 migration\n- 任何會刪資料或覆蓋工作的操作\n\n如果有人要求危險指令，請回覆：\n「這個指令可能造成資料遺失或覆蓋既有工作，請先確認後我再執行。」\n\n## 5) repo-skills\u002Fspec-driven-workflow.md\n\n# 技能：規格驅動開發\n\n在開始實作前，先產出：\n1. 簡短規格摘要\n2. 限制條件\n3. 假設\n4. 任務拆解\n5. 實作計畫\n\n工作流程：\n- Spec：先講要做什麼、為什麼做\n- Plan：再講怎麼做\n- Tasks：拆成可 review 的小步驟\n- Implementation：確認後才開始\n\n## 6) repo-skills\u002Fstack-specific.md\n\n# 技能：技術棧慣例\n\n把你團隊的慣例寫進來。\n\n範例：\n- TypeScript：公開函式盡量保留明確型別\n- Python：函式保持短小、好測試\n- React：避免不必要的 state\n- API：加上 request \u002F response 範例\n\n## 7) 範例起手式\n\n請使用這份 repo 的技能檔案，並遵守 spec-driven-workflow.md。\n這次任務請先：\n- 重述目標\n- 列出假設\n- 提出任務拆解\n- 等我確認後再開始實作\n\n## 8) 操作原則\n\n- 如果 AI 開始自作主張，先停。\n- 如果 AI 跳過測試，叫它回去補。\n- 如果 AI 輸出危險指令，先要求確認。\n- 如果 AI 說不出計畫，就不要讓它直接寫 code。\n\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>這份模板不花俏，因為我本來就不想要花俏。我想要的是一套能直接塞進 repo 的東西，讓 AI 別再自由發揮。這也是我從這份報告裡真正學到的事：大家不是在追一個更會聊天的工具，而是在找一種更穩的工作方式。\u003C\u002Fp>\u003Cp>如果你現在也被 AI 寫程式搞到有點火大，我會建議你不要先追新模型，先把技能文件補起來。挑一個最常出包的場景，先寫一份規則。寫完後你會很快發現，很多問題不是模型太笨，是你根本沒把邊界講清楚。\u003C\u002Fp>\u003Cp>來源網址：\u003Ca href=\"https:\u002F\u002Fdev.to\u002Fyanceyxin\u002Fgithub-weekly-trending-repositories-report-2g68\">DEV Community 的原始文章\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F\">star-history.com\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmattpocock\u002Fskills\">mattpocock\u002Fskills\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgithub\u002Fspec-kit\">github\u002Fspec-kit\u003C\u002Fa>。上面前半段是我根據原始報告整理的拆解，模板段落是我依照這些模式重新寫成可直接使用的版本。","我把 GitHub 趨勢報告拆成一套可直接複製的 AI 寫程式流程，重點是把提示詞變成技能與規則。","dev.to","https:\u002F\u002Fdev.to\u002Fyanceyxin\u002Fgithub-weekly-trending-repositories-report-2g68",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779134707718-sslr.png",[13,14,15,16,17],"AI agent","GitHub 趨勢","skills","spec-driven development","Claude Code","zh",0,false,"2026-05-18T20:04:26.488566+00:00","2026-05-18T20:04:26.45+00:00","done","cfcf732d-76b6-412b-82ab-8c04608b95a5","github-skills-repos-turn-ai-coding-into-workflows-zh","tools","664cda3c-65a0-4e8b-8342-62bdc48175ba","published",[30,31,32],"把 AI 提示詞改成 repo 裡可重用的技能文件。","先規格、再計畫、最後才實作，能少掉很多亂寫。","技能庫本質上是可執行的文件，不是花俏 prompt。","c3c88dd2-a940-438a-b359-0e5a24562273","[0.0065144035,0.010098843,0.029205058,-0.06694284,-0.008522839,-0.01808365,-0.01909781,0.0019816386,0.002893523,0.010740668,-0.022400219,-0.020650916,0.016512018,0.005463717,0.13614501,0.009152589,0.011010821,0.018421715,0.014978623,-0.009155305,0.010065306,0.03655866,0.00791641,-0.0030373465,-0.01652196,-0.01588933,0.0125342,0.043959703,0.051870853,-0.012705026,0.017050076,0.040690582,-0.00116367,0.007943155,0.014874507,0.010613528,-0.008212602,-0.0017743899,0.005773047,0.004240874,-0.019486805,0.022925815,-0.010109072,-0.004131056,0.011710003,-0.0035166987,-0.0030090965,-0.009652203,0.007577781,0.02722134,0.0082554445,0.009667598,-0.025148163,-0.15671828,0.020597626,0.02565884,-0.03767227,0.0075242016,0.007339885,0.0064475653,-0.0021415558,0.020513916,-0.025716817,-0.014083818,-0.0055991584,-0.016064674,0.03395473,0.0013045602,-0.0088895885,-0.014624399,-0.03727716,0.009165692,-0.006157839,-0.030496243,-0.00087024015,-0.0101222,0.00056022196,-0.02611617,-0.005136784,0.028090313,0.009251058,-0.022187797,0.011336347,0.0036559538,0.0034746649,-0.02613979,0.0033913709,0.0027166333,0.009920903,0.0060248044,0.020521997,0.035063602,-0.0038769448,-0.0068699345,0.008469278,-0.006151329,-0.009106189,0.0058104233,0.016772455,-0.02300125,-0.006208257,-0.034643218,0.018673893,-0.004070194,0.015239517,0.016032398,-0.011693226,-0.014874402,0.011743508,0.008521086,0.0202997,-0.027124966,-0.0023084334,0.013466681,0.013520814,-0.12652487,-0.010560262,0.00021609991,0.020828985,0.0212133,-0.020922597,0.007858457,0.0072581074,0.022075133,-0.010583341,-0.0151194185,0.0051077837,0.0048225583,-0.035065908,-0.019108202,-0.021997703,0.011499043,-0.0044869985,-0.005528661,-0.009279113,0.017677743,-0.011906997,-0.011640196,-0.003511008,-0.040148,-0.021050502,0.03590353,-0.023293309,-0.003133524,-0.025775855,-0.014190613,-0.017176924,0.022211468,0.001164836,-0.020172697,-0.0060047726,-0.02186439,-0.018254695,-0.012147456,0.009478478,-0.017276097,0.015627095,0.006456192,0.006626259,0.019181376,0.0047355476,-0.008172364,-0.012752997,0.016876051,0.008462151,0.033580195,0.0043827174,0.011522201,-0.0044540623,0.021766989,0.010800063,0.0027531611,0.021665573,0.008206087,0.012230517,-0.0042390577,-0.0070659714,-0.0111501515,0.009912972,-0.025526866,0.02507932,-0.0040387292,-0.0139117995,0.023075337,0.009817224,-0.023099637,-0.0037561073,0.0045811366,0.01705468,0.020340793,-0.018874861,0.0009582711,0.008574485,-0.0057282303,0.008803158,-0.0330228,-0.0028869675,-0.016732302,0.00044043688,0.029593283,0.0440932,0.007123784,0.0115763275,-0.041081823,0.032878976,-0.008302008,-0.015380685,-0.0077887354,0.016080359,-0.016338095,-0.042318482,0.011951239,0.024429582,0.0051187533,-0.01082559,-0.028067011,0.012502295,-0.0104842335,0.019679531,-0.016997665,0.0020291826,-0.004114232,0.0074049723,0.014575784,-0.02622269,-0.00611464,-0.0030628773,-0.010809448,-0.014798776,0.015555877,0.01519324,0.0013393817,-0.009462596,-0.0041044145,0.005066664,0.0017421113,-0.0017056757,0.01935599,0.0035065173,0.012225505,-0.017777357,0.010379576,0.016039263,0.0077012433,0.024046024,-0.025150906,-0.011114918,0.01115745,-0.008137771,0.0006465479,-0.0113383215,-0.016837012,0.012322377,-0.00456211,-0.018760566,-0.011106061,0.005885938,0.00894535,-0.013476335,0.002396003,-0.01920021,0.0013867138,-0.035016473,0.007517849,0.03150589,-0.0014759997,0.003354331,0.018676033,-0.0036322451,-0.028477695,-0.012996337,0.039567675,0.020320637,-0.0069901934,0.0053199525,0.019515004,-0.06430053,0.03484584,-0.007714651,-0.041585125,-0.008060829,0.014610812,0.01021217,-0.012268025,-0.026980244,0.00073242845,-0.014890157,-0.014467204,0.011940647,-0.013858933,0.0006590868,0.01615573,0.0036665797,-0.009900606,-0.0027250268,-0.004038675,-0.001200283,0.008986502,-0.0053083287,-0.003289168,0.006670536,-0.0127607435,0.034607027,0.04393861,-0.008560826,-0.0030071333,0.027157508,0.03548193,0.036853824,-0.020897608,-0.009062775,-0.0321082,-0.00015722215,-0.016961424,-0.0038261923,0.00079416856,0.0006389692,-0.019754425,0.011841652,-0.015034122,0.0007176641,-0.0046443962,0.002965583,0.01588767,-0.023223557,-0.00703302,-0.009628404,0.00477258,0.03370953,-0.010785525,0.01204002,0.03491828,0.031069528,-0.010975479,-0.0039931596,0.016666783,-0.013872125,0.0048943297,-0.020727873,0.0024605838,-0.022253836,-0.0017937056,-0.018486513,0.0187365,-0.023962732,0.019682586,0.014966146,-0.013593171,0.0013078174,-0.024117593,0.007985067,-0.015757535,0.0142663745,-0.003393765,-0.040314052,0.012541397,-0.012397988,-0.0035661424,0.028314304,0.0027940634,0.010599512,0.00688317,0.009595203,-0.012580679,0.016369823,0.013943205,0.007744245,0.021440152,0.0036328959,0.008405556,-0.010890118,-0.002069926,-0.011039414,-0.01521661,-0.0050805607,0.013545879,-0.018451834,-0.012181529,-0.013816925,-0.002184465,-0.007936261,0.014699815,-0.020747736,-0.01081063,-0.014450377,-0.011618851,0.012516646,0.0011015369,0.014314559,-0.012948998,0.033224545,-0.021342078,-0.009962799,0.019364921,0.012603806,-0.0027001454,-0.0044023194,-0.0052786823,-0.016267005,0.00048440567,0.006863681,0.004876808,-0.009276468,0.020825936,-0.0036229743,0.0010651031,-0.00429317,0.004553066,-0.018846957,0.0012222542,-0.00014464422,0.010262172,0.009527107,0.018075695,0.0123374155,0.006390321,-0.0008622628,0.0060405536,0.012266561,-0.029294636,0.0019467804,0.02904253,-0.0027070793,0.01269209,0.0044948864,-0.0073597473,0.0032017229,-0.013980362,0.0015258236,-0.025028555,0.005281556,-0.0071046497,0.009856961,-0.026459325,-0.035431553,0.00017774788,-0.013178644,0.0045412686,-0.028149327,-0.011012475,0.026834564,-0.01110616,0.0009008423,0.012564605,0.020934256,-0.009334488,0.006828285,-0.013212942,-0.042729303,-0.001849332,0.014580401,0.009621006,0.020234574,-0.0020334097,-0.010283467,-0.025724504,0.0003811948,0.03170016,0.0057315845,-0.021980992,-0.0030014082,0.009939926,-0.006023712,0.010565999,0.0011099356,-0.0041127605,0.015385164,-0.033839833,-0.009504385,0.03987501,-0.007632993,0.010494223,0.017026523,0.010025087,-0.012409835,-0.008228554,0.008050472,0.009793408,0.011800374,-0.028709421,-0.003804562,0.012791671,-0.0049213823,-0.0048715556,0.016277464,0.0024821528,-0.030019026,-0.008816328,-0.024753645,0.0037125738,0.02528377,0.027080895,-0.00073375524,-0.006666406,-0.03093247,-0.006230402,-0.009081546,-0.009614466,0.0031764237,-0.01003787,-0.0122349765,-0.01515802,0.019504033,-0.008196476,0.0038699752,0.008280524,-0.017091801,0.012709655,0.01048183,0.004599271,-0.011108653,-0.012405291,0.018356994,0.014003805,-0.018294508,-0.025487075,-0.009430181,0.010849709,-0.0007157203,-0.01554507,-0.0012037888,-0.0061465036,-0.0026636147,0.02607837,-0.011490892,-0.030783731,0.029072218,-0.014841314,0.004456491,-0.004569813,-0.013124066,0.022473425,-0.006102402,0.0017638007,0.01813423,-0.006625282,0.011705307,0.03417757,-0.01476862,-0.01626371,-0.0014341114,-0.0033705567,-0.01700515,0.009379783,0.049233217,-0.104865596,-0.014919267,0.010996815,-0.009406814,0.0128414715,-0.019498643,8.9580106e-05,-0.018048756,0.012769744,0.016497595,0.00598607,-0.033328805,0.0070358026,0.011380872,0.0017804762,-0.013999625,-0.018888716,-0.004516618,0.02722421,-0.037486598,-0.00080225844,-0.019621376,0.011866506,-0.02788875,-0.007607444,-0.0031998535,0.011614185,-0.008539176,-0.0028734207,0.0020118647,-0.03699869,0.005779697,0.017396014,0.012197124,-0.008770092,0.0064038327,0.016293531,-0.012735165,0.003259332,0.004442906,-0.004115256,0.0073828567,-0.024124121,-0.017919013,-0.014362311,-0.004220504,0.012654168,-0.009746712,0.012823884,0.0010424769,-0.061226152,-0.011032172,-0.0006592219,-0.023751188,0.033220377,0.0105139725,0.011970963,-0.0020542433,-0.031540215,0.019115455,-0.022701737,-0.0029338207,-0.0036617871,0.0017141144,-0.003978397,0.021388069,0.01983685,0.003821449,0.0058610877,0.009116796,0.010885226,-0.014768375,0.020945389,0.027092852,-0.009629704,0.019105569,-0.023299616,0.019607821,-0.017783297,0.00635544,-0.019523194,0.011522759,-0.07197615,-0.030633679,0.0069060693,0.0032017832,0.013263446,-0.008305923,0.028268218,-0.009438048,-0.018149177,-0.007663473,0.009124531,-0.010447103,-0.011999041,-0.023306884,0.015323759,0.010984365,0.00054809725,-0.00055731204,0.014830092,-0.010046342,-0.003346322,0.012715904,0.033451244,-0.023435932,0.008286232,0.0021791116,0.008511337,0.018775275,-0.0026466185,-0.025671482,-0.039485205,-0.1483848,-0.021051958,-0.010951188,-0.009707318,0.0062126718,0.019476041,0.011820979,0.02166917,0.0006714605,-0.015699603,0.006111471,-0.065775104,-0.03602302,-0.015494752,-0.0029734643,0.112387344,-0.004848272,-0.009882232,-0.022203939,0.00458298,0.019269522,-0.028205385,-0.03417084,-0.00048148457,-0.0018018564,-0.015885634,0.036021356,-0.0069771186,0.0015563959,0.0053140335,0.006449603,0.025170572,-0.018212616,-0.009640774,0.026093729,0.0006400165,-0.009864378,-0.015195875,0.021566665,0.035586093,0.009588053,0.02878379,-0.005282458,-0.0027862282,-0.031441454,0.02801461,-0.041081946,-0.022708325,-0.021933585,0.006105622,-0.024447937,-0.051310778,-0.007131098,-0.05791426,0.015711011,0.02163601,-0.013717311,0.017209573,0.029013388,-0.0045082807,0.022383546,0.012033731,-0.0049473057,0.020803541,-0.025520967,0.0025297522,0.041159905,0.027495673,0.0026230058,0.012211602,0.00014150381,0.025296899,-0.018056268,0.025449574,-0.025546536,-0.037336435,0.0017521688,0.030917523,0.017761337,-0.00037639932,0.004147222,0.00904501,0.024658978,-0.008622783,0.015522916,0.014990568,0.030477783,5.336999e-05,-0.008498101,-0.027294796,-0.01510977,0.010516644,0.009180188,0.025088478,0.02431778,0.027207619,-0.021156238,0.029351044,0.012597296,-0.0108394725,0.0013390097,-0.019244391,0.0072369124,-0.0141808465,0.031074656,0.008313209,-0.0043042595,0.0065283347,0.028205268,0.018211568]",{"tags":36,"relatedLang":46,"relatedPosts":50},[37,39,41,43,45],{"name":14,"slug":38},"github-趨勢",{"name":40,"slug":40},"claude-code",{"name":13,"slug":42},"ai-agent",{"name":44,"slug":44},"spec-driven-development",{"name":15,"slug":15},{"id":27,"slug":47,"title":48,"language":49},"github-skills-repos-turn-ai-coding-into-workflows-en","GitHub skills repos turn AI coding into workflows","en",[51,57,63,69,75,81],{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":26},"eb8ba5dd-3093-4c8e-8625-bfd5f11f3800","zhihu-soft-copywriting-four-tips-zh","知乎軟文教你寫出能賣貨的標題","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779146044192-aild.png","2026-05-18T23:13:42.179515+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":26},"8c6ebfe4-617a-4380-a06b-69b5a80bdde2","github-copilot-app-desktop-agent-preview-zh","GitHub Copilot 桌面代理預覽開放","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779141829436-mot7.png","2026-05-18T22:03:22.898714+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":26},"1e852f64-41ab-476c-83a6-d0c8732f86b1","xurrent-q2-2026-ai-release-turns-itsm-into-agents-zh","Xurrent Q2 AI 讓 ITSM 變成代理","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779140063861-o0jh.png","2026-05-18T21:33:56.788859+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":26},"6f625844-2519-4fac-8be2-d2d06a0686f5","top-ai-github-repositories-dominating-2026-zh","2026 最強 AI GitHub 倉庫盤點","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779136463477-0jml.png","2026-05-18T20:33:51.029803+00:00",{"id":76,"slug":77,"title":78,"cover_image":79,"image_url":79,"created_at":80,"category":26},"41688e9f-54e8-4270-b887-769090f99622","horizon-github-ai-news-briefings-zh","Horizon：GitHub AI 新聞簡報庫","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779135836565-0x4g.png","2026-05-18T20:23:28.561989+00:00",{"id":82,"slug":83,"title":84,"cover_image":85,"image_url":85,"created_at":86,"category":26},"7d325b37-35ff-4ecb-b9d4-a6cab8d0c48e","top-ai-prompt-engineering-tools-2026-zh","2026 必看 AI Prompt 工具整理","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779133443961-ijso.png","2026-05-18T19:43:34.399846+00:00",[88,93,98,103,108,113,118,123,128,133],{"id":89,"slug":90,"title":91,"created_at":92},"de769291-4574-4c46-a76d-772bd99e6ec9","googles-biggest-gemini-launches-in-2026-zh","Google 2026 最大 Gemini 盤點","2026-03-26T07:26:39.21072+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"855cd52f-6fab-46cc-a7c1-42195e8a0de4","surepath-real-time-mcp-policy-controls-zh","SurePath 推出即時 MCP 政策控管","2026-03-26T07:57:40.77233+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"9b19ab54-edef-4dbd-9ce4-a51e4bae4ebb","mcp-in-2026-the-ai-tool-layer-teams-use-zh","2026 年 MCP：團隊真的在用的 AI 工具層","2026-03-26T08:01:46.589694+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"af9c46c3-7a28-410b-9f04-32b3de30a68c","prompting-in-2026-what-actually-works-zh","2026 提示工程，真正有用的是什麼","2026-03-26T08:08:12.453028+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"05553086-6ed0-4758-81fd-6cab24b575e0","garry-tan-open-sources-claude-code-toolkit-zh","Garry Tan 開源 Claude Code 工具包","2026-03-26T08:26:20.068737+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"042a73a2-18a2-433d-9e8f-9802b9559aac","github-ai-projects-to-watch-in-2026-zh","2026 必看 20 個 GitHub AI 專案","2026-03-26T08:28:09.619964+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"a5f94120-ac0d-4483-9a8b-63590071ac6a","claude-code-vs-cursor-2026-zh","Claude Code 與 Cursor 深度對比：202…","2026-03-26T13:27:14.279193+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"0975afa1-e0c7-4130-a20d-d890eaed995e","practical-github-guide-learning-ml-2026-zh","2026 機器學習入門 GitHub 實用指南","2026-03-27T01:16:49.712576+00:00",{"id":129,"slug":130,"title":131,"created_at":132},"bfdb467a-290f-4a80-b3a9-6f081afb6dff","aiml-2026-student-ai-ml-lab-repo-review-zh","AIML-2026：像課綱的學生實驗 Repo","2026-03-27T01:21:51.467798+00:00",{"id":134,"slug":135,"title":136,"created_at":137},"80cabc3e-09fc-4ff5-8f07-b8d68f5ae545","ai-trending-github-repos-and-research-feeds-zh","AI Trending：把 AI 資源收成一張表","2026-03-27T01:31:35.262183+00:00"]