search-engine_skill

Overview

This skill helps you design and build production-grade search engines with robust indexing, retrieval, relevance controls, and evaluation workflows. It codifies architecture patterns, deterministic ingestion pipelines, staged retrieval, and safe rollout practices. The goal is predictable, testable search behavior that matches explicit retrieval contracts and business objectives.

How this skill works

The agent guides you through defining a retrieval contract (query types, latency, freshness, and error tolerance) and then proposes deterministic ingestion and indexing pipelines with stable identifiers and resumable jobs. It separates recall and precision into staged components (broad candidate retrieval, reranking, formatting) and prescribes versioned relevance policies, offline evaluation workflows, and idempotent index operations for safe rollouts. It also maintains local planning notes, experiment logs, and incident records to support continuous improvement and audits.

When to use it

• Designing a new search system for documentation, product catalogs, or knowledge bases
• Redesigning an existing search to improve relevance, latency, or scalability
• Scaling retrieval and ranking for growing or multilingual corpora
• Implementing safe rollout and rollback procedures for index and relevance changes
• Establishing repeatable evaluation and experiment pipelines before production writes

Best practices

• Start with a clear retrieval contract before choosing tools or vendors
• Implement ingestion as deterministic, repeatable stages with stable IDs
• Keep recall and precision separate: retrieve broad candidates, then rerank
• Version relevance policies and record feature weights and boosts
• Run offline benchmarks and comparisons before changing production behavior
• Build resumable, idempotent index operations and alias-based rollback plans

Example use cases

• Create a hybrid search for a product catalog supporting keyword and semantic queries
• Redesign enterprise document search to improve multilingual precision and filtering
• Add offline evaluation pipelines to validate relevance tuning before deployment
• Implement safe index updates for a high-availability search service using aliases
• Define retrieval SLAs and hit-quality benchmarks for a public-facing search API