deep-search

To perform deep web research, this skill runs targeted queries across multiple search engines like Bing, Google, and Baidu, opens relevant result pages, and collects browser snapshots and extracts to generate answers with numbered inline Markdown citations.

Multi-engine web search queries many engines like Bing, DuckDuckGo, and Baidu, switching when results are blocked or thin. You need it for current events, multi-region research, and fact synthesis when a single search engine lacks sufficient coverage.

You extract web evidence by opening promising search result links, waiting for content readiness, and collecting browser_snapshot or browser_extract outputs. These extracted sources are then tracked and tied to numbered inline Markdown citations for your final summary.

Yes, you can use browser automation to find current web evidence from dynamic pages. The process applies robust browsing steps with waits, snapshots, and extracts, while using fallback strategies and a retry ledger when pages are dynamic, blocked, or empty.

The best way to avoid redundant searches during multi-region research is using a de-duplication and retry ledger. This internal search ledger tracks previous queries and extractions, preventing repetitive tool actions and applying fallback strategies for blocked content.

Yes, the deep search approach works with non-English search engines like Sogou, 360, Baidu, and Quark. It supports multi-language research by running targeted queries across these regional engines and switching to alternatives if results are blocked or thin.