search-web_skill

Overview

This skill searches the web using Claude Sonnet's web_search tool and returns a single, pre-synthesized research Note. The Note contains a readable synthesis of findings plus a structured list of sources for follow-up. It’s designed to let the agent use a ready-made research summary rather than raw search results.

How this skill works

You provide a question-style query describing what you want to know; Claude Sonnet decides the search strategy, issues multiple searches with varied phrasings, reads and evaluates results, and synthesizes a single research Note. The Note includes a multi-paragraph synthesis, an attributed source list (url, domain, title, excerpt), a source count, the model used, and elapsed time. The skill requires a CLAUDE_API_KEY environment variable and returns a failure reason if the call fails or no results are found.

When to use it

• When you need a concise, readable research summary with source attributions.
• When you want Claude to plan and execute search strategies rather than crafting search-engine queries.
• When sources may be JavaScript-rendered or otherwise difficult to fetch with raw HTTP.
• When you need a single, immediately usable Note for downstream synthesis or extraction.
• When you want to quickly gather diverse perspectives (news, forums, papers) and a synthesized view.

Best practices

• Write the query as a question or decision-focused prompt describing what you want to learn, not as an instruction to search specific sites.
• Keep queries focused and specific (scope, timeframe, or required comparison) to get a tightly scoped synthesis.
• Use the returned Note directly for reading or use extract/project actions to pull specific fields like sources or key findings.
• Expect 15–45 seconds typical latency; allow up to the 120s timeout for complex topics.
• Check the sources list before relying on factual claims; use fetch-text for full-page content when needed.

Example use cases

• Compare recent developments and consensus on a software release or feature (e.g., FSD versions).
• Summarize latest trends and authoritative takes on a policy area (e.g., climate policy, AI governance).
• Collect practical differences and trade-offs between research methods (e.g., RLHF vs DPO).
• Gather and synthesize forum sentiment and expert commentary for product research.
• Produce a short research brief with cited URLs for a rapid decision meeting.