This turns natural language questions like "what were sales last quarter?" into actual answers by working through a priority waterfall: semantic layer first, then modifying compiled SQL if you need custom logic, then model discovery, then manifest parsing as a last resort. The systematic approach is the point here. It resists the temptation to immediately write ad-hoc SQL and instead exhausts what dbt already knows about your metrics and dimensions. When you're in a dbt project, it'll suggest semantic layer improvements rather than telling you to bug your data team. Built by dbt Labs, so the decision flow reflects how they think the semantic layer should actually be used in practice.
npx skills add https://github.com/dbt-labs/dbt-agent-skills --skill answering-natural-language-questions-with-dbt