【逆向WorkBuddy，开源一个搜索公众号文章的skill】

前段时间我拆解WorkBuddy自带的Deep Research功能的时候，发现在执行的过程中配备了一个可以搜索公众号文章的skill，叫做WeChat Article Search。

而且我发现，这个skill就这么直勾勾地躺在WorkBuddy的harness文件夹下面。

底层原理就是公众号文章给搜狗搜索开了一个口子，然后这个skill实际上就是通过调用搜狗的接口去爬文章。当然这里面的反爬和解析做了很多优雅的设计，技术大佬可以去学习一下～

开源地址：https://github.com/zjp1997720/wechat-article-search… 也可以使用 npx skills add zjp1997720/wechat-article-search 一键安装

zjp1997720/wechat-article-search

Source: https://github.com/zjp1997720/wechat-article-search

WeChat Article Search

中文文档

Search WeChat public account (公众号) articles by keyword. Returns structured JSON with title, summary, publish time, source account, and link — no API key required.

Agent Install

npx skills add zjp1997720/wechat-article-search -g -a codex --skill wechat-article-search -y

Requirements

• Node.js 18+ (tested on Node 20)
• One npm dependency: cheerio. Run npm install inside the skill folder after install.

What It Does

• Keyword search across WeChat public accounts via Sogou WeChat Search — returns 1–50 articles per query.
• Structured output: title, URL, summary, publish datetime, date text, relative time, source account name.
• Optional real-URL resolution (-r): converts Sogou redirect links into mp.weixin.qq.com direct links when reachable.
• Optional file output (-o): writes JSON to disk for archival.

How It Works

The skill drives a single self-contained Node.js script (scripts/search_wechat.js) that queries Sogou WeChat Search (weixin.sogou.com), parses the result HTML with cheerio, and emits structured JSON to stdout. No API keys, no MCP tools, no platform coupling — it works in any agent runtime that can run node.

Example Requests

Search WeChat articles about "Loop Engineering", give me 10.

搜一下"AI培训"相关的公众号文章，要 5 条，保存到 result.json。

CLI Usage

node scripts/search_wechat.js "<keyword>" [-n <count>] [-o <output.json>] [-r]

First run: npm install in the skill folder to pull cheerio.

Safety & Limits

• Uses public Sogou WeChat Search as the data source. Anti-bot rate limiting can intermittently return empty results or block URL resolution — retry with a different keyword or wait.
• -r URL resolution succeeds inconsistently under Sogou’s anti-spider policy. Failures keep the Sogou redirect link and set url_resolved: false.
• For study and research only. Do not use for large-scale commercial crawling. Excessive use may get your IP temporarily blocked.

Repository Layout

.
├── README.md
├── README.zh-CN.md
├── LICENSE
├── SKILL.md
├── agents/openai.yaml
├── package.json
└── scripts/
    └── search_wechat.js

License

MIT

智见AI-大鹏 (@zjp1997720): 【开源一个开源skill的skill】 看到这个标题你可能有点懵逼 功能就一句话：一键把你本地的skill开源到github，并且支持npx skills。 主打一个越分享越幸运！ 地址在这：