build-kb

What problem does it solve?

This Skill eliminates the manual, error-prone work of building and maintaining searchable, domain-aware knowledge bases for data projects, which is critical for ensuring LLM-generated SQL is accurate and aligned with business context.

Core Features & Use Cases

• Scoped or full-project generation: Build knowledge bases for entire projects or narrow them to specific files, tables, datasources, or business domains.
• Parallel domain exploration: Delegate read-only exploration of tables and documentation to subagents per business domain, classifying content into the correct storage types.
• Validated SQL corpus indexing: Enumerate every (natural language question, SQL) pair for reference with no curation required, and dual-route schema-non-inferable rules to knowledge stores for improved retrieval.
• Use Case: A data engineering team can use this Skill to generate a knowledge base for their e-commerce domain, including semantic models for orders and products tables, metrics for GMV and conversion rate, and all validated sales queries, so analysts can ask natural language questions and get accurate SQL instantly.

Quick Start

Use the build-kb skill to generate a vector knowledge base for the sales domain, including semantic models for orders tables, metrics for monthly revenue, and all validated sales queries from the queries/ directory.