websearch

name: websearch description: Guidance for web searching, page fetching, and source evaluation. trigger-servers: websearch

Web Search Skill

Workflow

• Start with a broad search, then narrow with follow-up queries if results are insufficient.
• Read the full page content for detailed questions — search snippets are often too brief.
• Prefer primary sources (official docs, repos, gov sites) over aggregators.
• If the user provides a URL, fetch it directly rather than searching for it.
• When you need to read 2+ URLs, use spawn_agent to read them in parallel — one sub-agent per URL. Do NOT call read_website sequentially when the reads are independent.

Quality

• Summarize findings in your own words — do not reproduce large blocks of source text.
• Note if information may be outdated and suggest the user verify time-sensitive details.