| name | web-search |
| description | Look something up on the web and read a page. Use when you need current information, a fact you're unsure of, or the contents of a specific URL. |
web-search
This skill needs no external binary — it uses the built-in web_fetch tool.
- To read a known page: call
web_fetch with its URL; it returns the page text (HTML stripped).
- To find something first: fetch a search results page, e.g.
web_fetch("https://duckduckgo.com/html/?q=<your+query>"), pick the most relevant result URL from
the text, then web_fetch that page.
Guidelines:
- Prefer primary sources. Quote or paraphrase accurately and tell the user where it came from.
- If a page is huge, fetch it and summarize the relevant part rather than dumping it.
- Don't claim something is current unless the page shows a date.