Mayf3 Smart Search

Problem

In agent workflows, web search is often unstable in three ways: one engine times out or returns empty results, Chinese and English queries are handled by the same backend inefficiently, and results are limited to summaries when full text is needed. mayf3-smart-search addresses this by connecting multi-engine search, language adaptation, fallback routing, and full-page extraction into one scriptable path.

How it works

• Local-first: it prefers the DDGS local library, which returns structured title, href, and body fields without manual HTML parsing.
• Language-aware routing: Chinese queries prefer bing; English queries prefer brave. --lang can force a language when needed.
• Automatic fallback: --fallback selects a backend chain by query language and retries on failure. If DDGS is unavailable, it can move to web_fetch targets such as Bing CN, 360, and Sogou.
• Search and extraction: --extract or extract() can fetch the full Markdown text of a result, which is useful for RAG, summarization, and source verification.
• Validation patterns: high-confidence tasks can combine brave + bing; WeChat content can use Sogou; quick emergency lookups can use Bing CN.

Boundaries

This skill is best used as a search and extraction tool, not as a truth verifier. Short, highly distinctive keywords work best; site:, quoted phrases, and after: can improve relevance. brave may return empty results under rapid consecutive requests, so switching to yandex or yahoo is a practical workaround. Newly published content may also be missing due to index latency.