generation

Generate cited answers by feeding retrieved documents into a synthesis model configured for multi-document context. The Skill automatically combines relevant information, attributes claims to source documents, and returns structured output with citations and metadata—eliminating manual answer assembly from document collections.

Answer generation includes fallback handling when retrieved documents don't match the query. The Skill manages low-context scenarios gracefully, exposing metadata about document count and relevance to help users understand when retrieval coverage is insufficient.

RAG (Retrieval-Augmented Generation) synthesis is designed for exactly this use case. The Skill operates as the final step in a RAG pipeline, consuming graded-relevant documents and producing evidence-backed answers with source attribution for technical documentation, specifications, or comparative analysis.

Source attribution happens automatically during synthesis. The Skill tracks which documents contributed to each claim, embeds citations in the answer text, and exposes document references and metadata in the output—so readers can verify claims against original sources.

Default configuration uses Claude Haiku 3.5 as the generation model, 0.3 temperature for consistency, and 2000 max tokens for output. These defaults balance response quality, cost, and length; they can be adjusted based on your synthesis complexity and latency requirements.

Multi-document synthesis reorganizes information into coherent narrative or structured format rather than preserving original document layout. The Skill extracts relevant claims across documents, integrates them logically, and rebuilds content with citations—ideal for comparative tables, summaries, or unified responses.