[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"category-skills-ru-data-analysis":3},{"items":4,"total":59,"page":60,"pageSize":61},[5,41],{"slug":6,"name":7,"title":8,"skillCategorySlug":9,"summary":10,"description":10,"skillCategoryName":11,"sourceUrl":12,"downloadUrl":13,"cliCommand":14,"authorName":15,"version":16,"license":17,"stars":18,"trendScore":19,"qualityScore":20,"recommendedTools":21,"recommendedToolNames":21,"supportedClients":23,"clients":27,"tags":38,"installHint":14,"useCases":39,"updatedAt":40},"sql-explain-agent","sql-explain","SQL Explain Agent","data-analysis","Explain SQL queries, suggest safer indexes, and produce database review notes.","Data analysis","https:\u002F\u002Fgithub.com\u002Fexample\u002Fagent-skills\u002Ftree\u002Fmain\u002Fsql-explain","https:\u002F\u002Fgithub.com\u002Fexample\u002Fagent-skills\u002Farchive\u002FHEAD.zip","npx skillfish add example\u002Fagent-skills\u002Fsql-explain","MCPMarkets","1.0.1","MIT",760,81,84,[22],"postgres-mcp",[24,25,26],"claude-desktop","cursor","cline",[28,32,35],{"slug":24,"name":29,"summary":30,"count":31},"Claude Desktop","Десктопный клиент с поддержкой локальной конфигурации MCP.",0,{"slug":25,"name":33,"summary":34,"count":31},"Cursor","AI-редактор кода, часто используемый с developer MCP.",{"slug":26,"name":36,"summary":37,"count":31},"Cline","Клиент для агентного кодинга с практичными MCP-интеграциями.",[9,17],[],"2026-06-17T00:00:00",{"slug":42,"name":43,"title":43,"skillCategorySlug":9,"summary":44,"description":45,"skillCategoryName":11,"sourceUrl":46,"downloadUrl":47,"cliCommand":48,"authorName":-1,"version":-1,"license":-1,"stars":49,"trendScore":50,"qualityScore":50,"recommendedTools":51,"recommendedToolNames":51,"supportedClients":52,"clients":53,"tags":56,"installHint":48,"useCases":57,"updatedAt":58},"alchaincyf-huashu-excel","alchaincyf\u002Fhuashu-excel","数据分析与 Excel 全流程 skill：体检脏表、清洗、对齐需求、分析、对账、交付。让 AI 算出来的数字经得起追问。跨 agent 通用，依赖仅 openpyxl。","—— 每个交过数字的人都被这么问过。 答不上来通常不是记性差，是那个数从一开始就没法被追溯。\n\n你写代码，错了会抛异常； 你做数据分析，错了什么都不会发生。一个 +161% 的错误， 交付出去的时候和正确答案长得完全一样。\n\n丢一份乱七八糟的 Excel 给它——标题占了第一行、表头两级、地区列用合并单元格、 金额带千分位、中间夹着「华东小计」、尾巴三行是「合计\u002F占比\u002F同比」。 主流做法算这份表的月度总额，。\n\n这个 skill 存在的全部理由，就是让那句反问能被答上来。它先读原始单元格再动 pandas，把表里的「合计」行当成免费的校验和拿来对账，交付前跑一遍 master check——不通过就不给你数字。\n\n跨 agent 通用——Claude Code、Cursor、Codex、OpenClaw、Hermes 都能装。\n\n发现表头位置不对就加 header=，发现数字读不出来就清掉千分位。看起来已经挺周全了。\n\n- 合计 行的 15,368,317 被当成了一家门店 - 华东小计 的 6,157,150 又被当成一家门店 - 有一行是粘贴事故造成的完全重复，算了两次\n\nPanko (1998) 的研究说 。欧洲电子表格风险兴趣组 （EuSpRIG）从 1995 年起持续收录见诸媒体的事故——最著名的是 Reinhart-Rogoff 论文，Excel 选区少选了 5 行，把 +2.2% 的 GDP 增速算成 −0.1%， 而那篇论文当时是全球紧缩政策的主要学术依据。","https:\u002F\u002Fgithub.com\u002Falchaincyf\u002Fhuashu-excel","https:\u002F\u002Fgithub.com\u002Falchaincyf\u002Fhuashu-excel\u002Farchive\u002FHEAD.zip","npx skillfish add alchaincyf\u002Fhuashu-excel",377,41,[],[24,25],[54,55],{"slug":24,"name":29,"summary":30,"count":31},{"slug":25,"name":33,"summary":34,"count":31},[9],[],"2026-09-10T03:23:40.120001",2,1,24]