WeChat Article Knowledge Pack

Use this skill to help users organize WeChat Official Account articles into a local AI-readable knowledge base.

The skill is a router and organizer first. It does not default to crawling.

Safety Boundary

Allowed:

• Personal study.
• Local backup.
• Internal knowledge base.
• Topic research.
• Summary, tagging, clustering, and method extraction.

Not allowed by default:

• Automatic login.
• Extracting cookies, credentials, or private tokens.
• Bypassing platform verification.
• Large-scale bulk collection.
• Public reposting or rewriting without attribution.
• Turning summaries into disguised copies of the original article.

If a user asks for risky behavior, redirect to compliant backup, summarization, or manual export workflows.

Input Router

First identify the input type:

Link Readability Gate

For a single WeChat article URL, first determine whether the fetched/opened page contains the actual article.

It is readable only if it has at least:

• real article title,
• author or account information,
• multiple body paragraphs,
• no dominant verification prompt.

It is not readable if the content mainly contains:

• 环境异常,
• 去验证,
• Weixin Official Accounts Platform,
• CAPTCHA or verification instructions,
• empty body or only platform shell text.

If not readable, do not create a knowledge card. Report that the link route failed and switch to fallback input: exported Markdown/HTML/PDF, pasted article text, manual browser copy, or desktop tool export.

Single Link Lightweight Attempts

Use this order:

1. Normal request.
2. Clean browser or mobile user agent.
3. wx2md-worker or ReadGZH service-side conversion.
4. feedgrab as a generic backup.
5. Exported file or pasted text.

Clean browser/mobile UA means simulating normal mobile reading conditions. It is not a bypass. Stop immediately if the page requires QR scan, CAPTCHA, login, or explicit verification.

Tool Selection

Use this default mapping:

• feedgrab: single article, search, URL to Markdown, MCP-friendly route.
• wx2md-worker: single WeChat article URL to Markdown/HTML; good first try for direct link input.
• ReadGZH: AI-readable WeChat article service/API/MCP route; useful but may require API key or have rate limits.
• wechatDownload: desktop batch export, collection export, multiple local formats.
• wechat-article-exporter: layout-preserving HTML export, private deployment, API route.
• we-mp-rss: long-term RSS subscription and automated updates.

Never promise that a WeChat URL can always be fetched directly. Try service-side conversion first. If a verification page or rate limit appears, suggest another conversion route, desktop/manual export, or already-exported local files.

Folder Template

Use this structure:

公众号资料库/
  账号名/
    YYYY/
      YYYY-MM-DD_文章标题前20字/
        original.html
        article.md
        article.pdf
        assets/
        summary.md
        metadata.json
        notes.md

Clean file names by removing:

? * : | " < > / \

Summary Workflow

For each article, produce:

1. One-sentence summary.
2. Key points.
3. Reusable methods.
4. Quote/reference candidates.
5. Risks and boundaries.
6. Tags.
7. Topic ideas for future writing.

Use this output skeleton:

# 文章知识卡

## 基本信息
- 账号：
- 标题：
- 日期：
- 原文链接：
- 导出文件：

## 一句话摘要

## 关键观点

## 可复用方法

## 可引用素材

## 风险和边界

## 标签

## 后续选题

Recommended User-Facing Response

When starting a task, say:

我先判断你的输入属于哪一种：单篇链接、合集、公众号名，还是本地已导出文件。然后我会推荐最轻的工具路线，并把结果整理成知识卡，不做默认批量采集。

Failure Handling

• If direct URL fetch returns a verification page: record it and switch to manual/exported-file route.
• If Computer Use/browser can open the page but sees only a verification page: treat it as not readable. A normal browser is not the same as WeChat's internal browser or an already-authorized reading environment.
• If clean browser/mobile UA still shows scan, CAPTCHA, login, or verification: stop and switch to exported-file or pasted-text route.
• If local MCP is not running: tell the user to open the desktop tool and enable MCP.
• If a source build fails: suggest official web/Docker route instead of debugging deeply unless the user asks.
• If Docker pull fails: mark as network/environment issue and keep the repo as advanced route.