Web search
Skill kohoj/skills/web-search
Claude Code skills for web scraping & content extraction — Twitter/X tweet scraper, Xiaohongshu (Little Red Book) OCR to Markdown. CDP + httpx architecture.
npx -y skills add kohoj/skills --skill web-searchAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Search the web via DuckDuckGo. Use when you need to find current information, discover URLs for deeper reading, or research a topic beyond your training data.
SKILL.md
1.5 KB, as published. Nobody here has run it
Web Search
Your knowledge has a cutoff. The world does not. When a question requires current information — prices, events, announcements, regulations, people — search before answering.
Search finds URLs. It does not read them. When a result looks relevant, follow up with WebFetch to read the full page. Search and read are separate acts — keep them separate.
Usage
.agents/skills/web-search/search "query"
| Flag | Effect |
|---|---|
--max N | Number of results (default: 5) |
--news | Search news instead of web (includes source and date) |
--region xx-xx | Region bias (e.g. cn-zh, us-en, hk-tzh) |
Examples
.agents/skills/web-search/search "TSMC N2 production timeline 2025"
.agents/skills/web-search/search "semiconductor export controls" --news
.agents/skills/web-search/search "美联储利率决议" --max 10 --region cn-zh
When to search
- User asks about something that changes (prices, policies, people, events)
- You need to verify a claim or find a primary source
- You are building a view and need current evidence
- The heartbeat flags a thread worth following
When not to search
- The answer is in the workspace (check WORLDVIEW.md, views/, journal.jsonl first)
- The question is conceptual or analytical — search adds noise, not signal
- You already have a URL — use WebFetch directly