Web Content Fetcher (WeChat images fix)

该 Skill 接收网页 URL，调用自带 Python 脚本抓取并提取正文，默认快速 HTTP 模式，内容过少时自动转为无头浏览器 stealth 模式；也可按域名表指定模式，或用 Jina Reader 兜底。输出保留标题、链接、图片、列表和代码块的 Markdown，并可选 JSON 元数据；不负责离线下载图片或内容总结。

haanya168

热度 84.61

最近更新 2026-05-11 17:35:23

分类：搜索依赖：代码执行 / 联网浏览搜索 / 第三方 API应用场景：研究 / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add web-content-fetcher-hanya

Metadata

name: web-content-fetcher
description: Extract article content from any URL as clean Markdown. Uses Scrapling script as primary method (with auto fast→stealth fallback), Jina Reader as alternative for simple pages. Preserves headings, links, images, lists, and code blocks. Use this skill whenever the user wants to fetch, read, extract, scrape, or summarize content from a URL — including blog posts, news articles, WeChat articles (微信公众号), documentation pages, or any web page. Also trigger when the user says things like "帮我读一下这篇文章", "抓取这个网页", "提取正文", or "read this page for me".

Web Content Fetcher

Given a URL, return its main content as clean Markdown — headings, links, images, lists, code blocks all preserved.

Note: This skill extracts content + remote image URLs. If the user wants an "offline" copy (download images to local disk and rewrite links), add a post-processing step (not included by default in this skill).

Extraction Strategy

Always try one method per URL — don't cascade blindly. Pick the right one upfront.

URL
 │
 ├─ 1. Scrapling script (preferred)
 │     Run fetch.py — check the domain routing table to decide fast vs --stealth.
 │     Works for most sites. Returns clean Markdown directly.
 │
 └─ 2. Jina Reader (fallback — only if Scrapling fails or dependencies not installed)
       web_fetch("https://r.jina.ai/<url>")
       Free tier: 200 req/day. Fast (~1-2s), good Markdown output.
       Does NOT work for: WeChat (403), some Chinese platforms.

Scrapling script

python3 <SKILL_DIR>/scripts/fetch.py "<url>" [max_chars] [--stealth]

<SKILL_DIR> is the directory where this SKILL.md lives. Resolve it before calling the script.

The script has two modes built in:

Default (fast): HTTP fetch, ~1-3s, works for most sites
--stealth: Headless browser, ~5-15s, for JS-rendered or anti-scraping sites

When run without --stealth, the script automatically falls back to stealth if the fast result has too little content. So you rarely need to specify --stealth manually — the only reason to force it is when you already know the site needs it (see routing table), which saves the initial fast attempt.

Domain Routing

Use this table to pick the right mode on the first call:

Domain	Command	Why
`mp.weixin.qq.com`	`fetch.py <url> --stealth`	JS-rendered content
`zhuanlan.zhihu.com`	`fetch.py <url> --stealth`	Anti-scraping + JS
`juejin.cn`	`fetch.py <url> --stealth`	JS-rendered SPA
`sspai.com`	`fetch.py <url>`	Static HTML
`blog.csdn.net`	`fetch.py <url>`	Static HTML
`ruanyifeng.com`	`fetch.py <url>`	Static blog
`openai.com`	`fetch.py <url>`	Static HTML
`blog.google`	`fetch.py <url>`	Static HTML
Everything else	`fetch.py <url>`	Auto-fallback handles it

Script Options

# Basic — auto-selects fast or stealth
python3 <SKILL_DIR>/scripts/fetch.py "https://sspai.com/post/73145"

# Force stealth for known JS-heavy sites
python3 <SKILL_DIR>/scripts/fetch.py "https://mp.weixin.qq.com/s/xxx" --stealth

# Limit output to 15000 characters (default: 30000)
python3 <SKILL_DIR>/scripts/fetch.py "https://example.com/article" 15000

# JSON output with metadata (url, mode, selector, content_length)
python3 <SKILL_DIR>/scripts/fetch.py "https://example.com" --json

Install Dependencies

First use only — the script checks and tells you if anything is missing:

pip install scrapling html2text

If on system-managed Python (macOS/Linux), add --break-system-packages or use a venv.

Failure Rules

Same URL fails once → give up, tell the user "unable to extract content from this URL"
Do not retry — each failed call wastes context tokens

WeChat-specific gotchas

WeChat often uses lazy-loaded images where the real URL is in data-src and src is a tiny placeholder.
The extractor script normalizes these to real src URLs before running html2text.
If you ever see Markdown image lines that contain weird URL-encoded SVG fragments (e.g. ...%3Csvg...%3E) appended after the closing ) of an image, it means the placeholder leaked into Markdown parsing; update fix_lazy_images() in scripts/fetch.py to remove/replace placeholder data:image/svg+xml src values.