performing-searches_skill

Overview

This skill provides concurrent web and code search capabilities with hybrid retrieval and embedding re-ranking to improve relevance. It supports multi-keyword concurrent queries (up to three), URL content extraction with simple JS rendering fallback, and library/documentation search for programming projects. Use it to augment agents that need fast, relevant search results across web, code libraries, or local search feeds.

How this skill works

The skill queries a SearXNG instance for raw search results, optionally applies an embedding-based re-ranking step to boost relevance to around 80%, and performs link deduplication and hybrid retrieval across multiple result pages. It exposes endpoints for concurrent search, URL reading (with pagination and heading extraction), library search, and library docs lookup, and can be run via Docker or an npm package.

When to use it

• When an agent must search the web and return concise, relevant results quickly.
• When you need to look up programming documentation, examples, or library APIs.
• When local network search or a self-hosted SearXNG instance is required for privacy or control.
• When you want hybrid retrieval with embedding re-ranking to improve result relevance.
• When performing parallel searches across multiple keywords or domains (limit 3).

Best practices

• Provide a running SearXNG instance and set SEARXNG_URL before using the skill.
• Limit concurrent searchedKeywords to 3 to avoid timeouts and lower noise.
• Enable an embedding endpoint (EMBEDDING_BASE_URL) for hybrid retrieval when higher relevance is needed.
• Increase SEARCH_TIMEOUT_MS for multi-page or hybrid retrieval (30–60s recommended).
• Filter out video-heavy sources in SearXNG to improve result quality for textual queries.
• Run via Docker for the easiest and most reproducible deployment.

Example use cases

• Agent performs concurrent searches for several candidate queries and ranks results for the next reasoning step.
• Automated lookup of React or other library docs and code examples using library_search and library_docs.
• Scraping and extracting page content for summarization or question answering via the read endpoint.
• Self-hosted private search for agents operating in air-gapped or regulated environments using SearXNG.
• Batch testing of search relevance across different query formulations with embedding re-ranking enabled.