| name | all-net-search-read |
| description | Public social media content search and extraction tool. Supports WeChat Official Accounts, Xiaohongshu public web results, Twitter/X public profiles, YouTube, Reddit, Bilibili public metadata/transcripts, and Weibo visible posts. Triggers: search WeChat, search Xiaohongshu, search Weibo, search Twitter, search Bilibili, search YouTube, search Reddit, read article, full-web search, podcast transcript. Not for: internal documents; login-required or premium content; cookies/tokens/SMS/QR login; large-scale systematic scraping. |
🕵️ All-Net Search & Read 18.0.0
Give your AI Agent the ability to search and read across all major platforms — WeChat, Weibo, Xiaohongshu, Twitter/X, YouTube, Reddit, Bilibili, and more.
⚠️ Gotchas
Known pitfalls — check here first before debugging.
⚠️ Xiaohongshu full note pages often require login → Do not request or store cookies, SMS codes, QR login, or session tokens. Use public web search results only; if the user needs exact note text, ask them to paste the content or provide a public page that is readable without login.
⚠️ Platform-specific search → Each platform uses its own search strategy (site: prefix + web search). Don't mix platform search methods.
⚠️ Twitter/X profile + timeline don't require login → Use agent-reach (xreach) for public profiles and timelines. Search functionality is limited without authentication; do not request auth tokens or cookies.
⚠️ Web search backend → Supports two tiers: agent-reach (xreach/xread) for best results across 14+ platforms, with Jina Search API (free, no key) as automatic fallback. Install agent-reach via npx clawhub install agent-reach for full capabilities.
⚠️ 小红书沙箱 IP 被封或页面要求登录 → 只能用 web search site:xiaohongshu.com 降级获取公开摘要或第三方索引结果。
⚠️ 微信公众号需要代理 → 沙箱直连微信服务器超时，必须配置 HTTP 代理（HTTP_PROXY）。
⚠️ B站字幕/弹幕可能需要登录 → 不请求 cookie；仅处理无需登录即可访问的公开视频元数据、音频或转写结果。
⚠️ 多链接批量获取需确认 → 一次请求含多个 URL 时，必须列出并获得用户确认（最多 5 个/批）。
🛑 Hard Stop
If the same tool call fails more than 3 times, stop immediately. List all failed approaches and reasons, mark as "Manual intervention required", and wait for user confirmation.
Input / Output Contract
Required input is one of: a public URL, a platform + keyword query, a public account/profile identifier, or pasted content. For multi-link requests, list the URLs and obtain confirmation before processing more than one link.
Return a fixed result shape:
- Task interpreted: platform, query/URL, and access mode.
- Retrieval path: primary tool and any fallback used.
- Results: extracted public content, summaries, metadata, or transcript snippets.
- Source evidence: links, timestamps, account/profile identifiers, and caveats.
- Missing or blocked content: login walls, deleted content, rate limits, or unsupported private access.
Do not output private content, cookies, tokens, login instructions, or claims that a login-only page was fully read.
✨ Core Features
| Feature | Description |
|---|
| 🔍 Full-Web Search | Semantic search across the internet with time/source filtering |
| 📱 WeChat Articles | Search and read WeChat Official Account articles |
| 🐦 Twitter/X | View tweets, user profiles, search topics |
| 📕 Xiaohongshu | Public web-search snippets and source-page discovery |
| 🎬 Bilibili | Public video info and transcript-oriented workflows |
| ▶️ YouTube | Video info, subtitles, comments |
| 💬 Reddit | Post and comment extraction |
| 📝 Web Distilling | Convert any webpage to clean Markdown |
| 📝 Content Summary | Auto-summarize long articles |
| 🏷️ Keyword Extraction | Auto-extract key info, people, terms |
| 🔔 Scheduled Monitoring | Monitor specific keywords/accounts for updates |
| 📥 Multi-link Fetching | Batch fetch multiple public links (max 5 per request, user confirms each) |
| 📚 Bookmarks | Save interesting content |
| 🕐 History | Remember previous searches |
🎯 Trigger Words
Search
搜 {keyword} — Full-web search (e.g., 搜 OpenAI latest news){keyword} recent news — Filter by time{platform} {keyword} — Platform-specific search
Platform-Specific
看 {account name} 最新文章 — WeChat article reading搜小红书 {keyword} — Xiaohongshu search搜 Twitter {keyword} / 搜推特 {keyword} — Twitter search搜 B站 {keyword} — Bilibili search搜 YouTube {keyword} — YouTube search搜 Reddit {keyword} — Reddit search
Content Processing
{URL} — Web distilling/reading总结 {content/link} — Content summary提取关键词 {content/link} — Keyword extraction
Management
我的收藏 — View bookmarks收藏这个 — Bookmark current content搜索历史 / 我的记录 — History监控 {keyword} — Set up scheduled monitoring
📖 Usage Examples
1. Full-Web Search
搜 OpenAI latest funding news
→ Returns relevant content from across the web
2. WeChat Article Reading
搜 机器之心 最近文章
→ Returns latest article list from the WeChat account
3. Platform-Specific Search
小红书 AI tool recommendations
推特 OpenAI latest updates
B站 deep learning tutorials
4. Web Distilling
https://example.com/article
→ Auto-converts to clean Markdown, removes ads
5. Content Summary
总结 https://mp.weixin.qq.com/s/xxx
→ Returns core takeaways
6. Keyword Extraction
提取关键词 https://news.ycombinator.com/xxx
→ Returns key people, terms, organizations
7. Scheduled Monitoring
监控 AI Agent updates
→ Push daily/weekly updates on the topic
8. Bookmark Management
我的收藏
收藏这个
🛠️ Technical Backend
Each platform uses appropriate tools:
| Platform | Backend Tool | Account Required |
|---|
| Xiaohongshu | Public web search fallback only | ❌ No; login-only notes are out of scope |
| Weibo | Web search + Playwright visitor-mode public rendering | ❌ No |
| Bilibili | Public Bilibili API + yt-dlp where permitted | ❌ No |
| YouTube | yt-dlp | ❌ No |
| Reddit | Reddit JSON API | ❌ No |
| Twitter/X | xreach (agent-reach) → web search fallback | ❌ No; authenticated search is out of scope |
| General Web | xread (agent-reach) → Jina Reader / curl fallback | ❌ No |
⚠️ Important: Xiaohongshu search and note detail pages often require login. This skill does not request or store cookies; use public search snippets, public source pages, or user-provided text only.
🔄 Execution Strategy (Fallback Rules)
Xiaohongshu Search
Xiaohongshu full note pages often require login. This skill does not collect login cookies or run login flows.
For the public-only fallback policy, see references/xiaohongshu-public-fallback.md.
Quick path: public web search (site:xiaohongshu.com) → readable public result → summarize with source caveat; login wall → ask user to paste content or provide a readable public source.
Twitter/X Search / Profile / Timeline
Step 1: Use agent-reach (xreach) for profile or timeline (no login required)
xreach twitter profile <handle>
xreach twitter timeline <handle> --count 20
✅ Success → Return followers, bio, tweet list
❌ agent-reach unavailable / API change → Step 2
Step 2: Fallback web search
Web search for "site:twitter.com {keyword}"
✅ Success → Return tweets
❌ Failure → Inform user to configure search API or try again later
⚠️ In environments where direct X access is blocked, ensure HTTP_PROXY is set for agent-reach to connect.
General Fallback Principle
Prefer agent-reach (xreach/xread) → platform-specific API → Jina web search → curl
First Use
Run the dependency check script before first use:
bash scripts/setup.sh
The agent will auto-run this on first trigger; usually no manual action needed.
📦 Dependencies & Installation
Python packages
- Python 3.8+
requestsbeautifulsoup4
Optional: Search API
For better search results, configure a search API key in config/default_config.json. Supported: Brave Search, SerpAPI, or any compatible web search API.
Recommended: Agent Reach (enhanced multi-platform search & read)
agent-reach provides powerful multi-platform search and content extraction across 14+ platforms. The skill auto-detects and uses it when available; without it, falls back to Jina Search API.
Install agent-reach (choose one):
pip install agent-reach
agent-reach install --env=auto --safe
npx clawhub install agent-reach
After installation, run agent-reach doctor to check which platforms are available. The skill automatically detects xreach (search) and xread (content extraction) CLIs at runtime.
Optional add-ons for agent-reach (unlock more platforms):
npm install -g mcporter
mcporter config add exa https://mcp.exa.ai/mcp
pip install 'rdt-cli>=0.4.2'
Without agent-reach, the skill falls back to Jina Search API (free, no key required) — still works for most use cases.
Optional: Dependency Skills
- agent-reach — Multi-platform search & read (xreach/xread/mcporter CLIs)
各平台详细技术笔记（代理配置、API 特性、已知限制）见 references/platform-notes.md。
🔒 Security Constraints (Mandatory)
- Public internet content only: Do not access internal/corporate domains
- Multi-link fetching requires user confirmation: When multiple URLs are in one request, agent must list them and get explicit user approval before fetching (max 5 per batch)
- No large-scale scraping: Do not use for systematic batch collection, automated loop downloads, or high-frequency requests to a single site
- Respect robots.txt and anti-scraping protections
- Scheduled monitoring requires user notification: Before setting up monitoring tasks, explain monitoring target, frequency, and data destination to the user
- No credential collection: Do not ask for cookies, tokens, SMS codes, QR-code login, browser profiles, or private account state.
Changelog
| 版本 | 日期 | 变更 |
|---|
| 18.0.0 | 2026-04-15 | Quality fix: standardize versioning, add Gotchas/Hard Stop sections |
