Ink Reader

Intelligently read any URL content. Auto-detect platform, pick the best fetch strategy, output clean Markdown.

When to Activate

Activate this skill when the user:

• Shares a URL and asks to read / fetch / view / grab its content
• Says "read this link", "what does this say", "fetch this article"
• Uses /ink-reader <url>

Fetch Strategy Overview

Three-layer fallback, zero configuration required:

Layer 1: Jina Reader    → Free, no API key, covers most public content
Layer 2: WebFetch        → Claude Code built-in, direct URL reading
Layer 3: Playwright MCP  → Browser automation, handles login-required sites

Platform Detection

Match the URL domain to determine platform and strategy routing:

Platform	Domain Contains	Needs Login	Strategy Order
WeChat	mp.weixin.qq.com	Yes	Jina → Playwright
Zhihu	zhihu.com	No	Jina → WebFetch
Bilibili	bilibili.com, b23.tv	No	Jina → WebFetch
Toutiao	toutiao.com	No	Jina → WebFetch
Weibo	weibo.com, m.weibo.cn	Yes	Jina → Playwright
Xiaohongshu	xiaohongshu.com	Yes	Jina → Playwright
Douyin	douyin.com	No	Jina → WebFetch
X/Twitter	x.com, twitter.com	Partial	See X/Twitter Flow
Generic	anything else	No	Jina → WebFetch

Routing Rules

• No login required: Jina → WebFetch → Playwright MCP (if available)
• Login required (WeChat, Weibo, Xiaohongshu): Jina → Playwright MCP (skip WebFetch, it won't help)
• X/Twitter: Dedicated flow below

Execution Steps

Step 1: Identify Platform

Parse the URL domain and match against the platform table above.

Step 2: Fetch Content

For normal platforms (no login needed)

1. Try Jina Reader:

   • Use WebFetch with URL: https://r.jina.ai/{original_url}
   • Prompt: "Extract the article title, author, publish time, and full body content. Return as-is in Markdown."
   • If result is meaningful (> 100 chars, no verification page), use it.

2. Try WebFetch direct:

   • Use WebFetch with the original URL directly.
   • Prompt: "Extract the article title, author, publish time, and full body content."
   • If result is meaningful, use it.

3. Try Playwright MCP (if available):

   • Navigate to the original URL.
   • Wait for content to load.
   • Take a snapshot and extract content.
   • If Playwright MCP is not available, skip this step.

For login-required platforms (WeChat, Weibo, Xiaohongshu)

1. Try Jina Reader (same as above, sometimes works even for login-required sites).

2. Try Playwright MCP (if available):

   • Navigate to the original URL.
   • If a verification/login page is detected, inform the user.
   • If Playwright MCP is not available, inform the user:

     "This platform requires login. Install Playwright MCP to enable browser-based reading."

For X/Twitter

1. Extract status ID from URL:

   • Pattern: x.com/{user}/status/{id} or twitter.com/{user}/status/{id}
   • Strip query parameters.

2. Try Thread Reader App via Jina:

   • Use WebFetch with URL: https://r.jina.ai/https://threadreaderapp.com/thread/{status_id}.html
   • Prompt: "Extract the full thread content including all tweets. Return in Markdown."
   • If result is meaningful (> 100 chars, contains actual thread content), output as thread.

3. Try Jina on original X URL:

   • Use WebFetch with URL: https://r.jina.ai/{original_url}
   • Prompt: "Extract the tweet content, author, and timestamp."
   • If result is meaningful, output as single post.

4. Try Playwright MCP (if available):

   • Navigate to https://threadreaderapp.com/thread/{status_id}.html
   • Extract content from the page.

Step 3: Validate Content

Content is valid when ALL of these are true:

• Length > 100 characters after trimming
• Does NOT contain these verification markers: "环境异常", "完成验证", "请完成验证", "access denied", "please verify"
• Is NOT a login wall or CAPTCHA page

If content fails validation, treat it as a failure and try the next strategy.

Step 4: Output

Use the output format specified below.

Output Format

Success

# {Title}

**Source**: {Platform Name}
**Author**: {Author name, omit if unavailable}
**Published**: {Time, omit if unavailable}
**URL**: {Original URL}
**Strategy**: {Jina Reader / WebFetch / Playwright MCP}

---

{Body content in Markdown}

Rules:

• Only include Author and Published lines if the information is actually available.
• Do NOT fabricate metadata. If it's not in the fetched content, omit it.
• Keep images as remote URLs. Do NOT attempt to download images.
• Clean up excessive whitespace, navigation elements, ads, and cookie banners from the content.

Failure

# Failed to read URL

**URL**: {url}
**Platform**: {detected platform}
**Attempted strategies**:

- {strategy 1}: {error reason}
- {strategy 2}: {error reason}

**Suggestions**:

- {contextual suggestions}

Contextual suggestions by scenario:

• Login-required platform + no Playwright → "Install Playwright MCP to enable browser-based reading for this platform."
• WeChat verification → "WeChat has strict anti-scraping. Try opening the link in a browser and copying the content manually."
• All strategies returned empty → "The page may require JavaScript rendering. Try using Playwright MCP."
• X/Twitter thread failed → "Try opening https://threadreaderapp.com/thread/{id}.html in your browser."

Save Mode

When the user says "save", "save it", "keep this", or "save to file" AFTER a successful read:

1. Create directory ./ink-reader-clips/ in current working directory (if not exists).
2. Write file: ./ink-reader-clips/{YYYY-MM-DD}_{sanitized_title}.md
3. File content:

---
title: "{Title}"
source: "{Platform Name}"
url: "{Original URL}"
saved_at: "{YYYY-MM-DD HH:MM:SS}"
---

{Body content}

1. Sanitize title for filename: remove <>:"/\|?*, replace whitespace with -, truncate to 50 chars.
2. Report: "Saved to ./ink-reader-clips/{filename}"

Do NOT auto-save. Only save when explicitly asked.

Important Notes

• No Python scripts: Everything is done through Claude Code's built-in tools.
• No API keys needed: Jina Reader is free and keyless.
• Playwright MCP is optional: The skill works without it, just with reduced capability for login-required platforms.
• Images stay remote: Never download images. Keep original URLs in the Markdown output.
• Respect content: Output the content faithfully. Do not summarize or modify unless the user explicitly asks.