Realtime Web Search

**Quick judgment:** A focused, lightweight skill that turns the model into a real-time web search assistant using `web_search_preview`, well-suited for freshness-critical queries but dependent on tool availability and base-model safety. **Strengths:** - Enforces use of live web search first, maximizing recency. - Encourages authoritative sources and clear summarization. - Handles lack of tool support with explicit fallback messaging instead of silent failure. - Automatically replies in the user’s language, improving UX. **Key risks / limitations:** - Hard dependency on `web_search_preview`; on models/environments without this tool, the skill loses its main value and falls back to static knowledge. - No explicit safety or misinformation-mitigation instructions beyond “prefer authoritative sources,” so harm prevention relies almost entirely on the underlying model. - Limited control over search behavior (no parameters like region, time window, or source preferences), and no explicit strategies for empty/low-quality results. **Recommended scenarios:** - Agents or workflows that need **recent information** on news, technology, products, or services and have reliable access to `web_search_preview`. - As a drop-in “freshness booster” in larger assistants that otherwise rely on static training data. - Not ideal for deployments without guaranteed web-search tooling or where strong, explicit safety/misinformation controls are required at the skill level.