dbt Analytics Engineering AI Skill Guide

Overview & Engine Architecture

dbt compiles Jinja-SQL models into warehouse DDL/DML, managed via a DAG of ref() / source() edges. Runs materialize views/tables/incremental models; tests and documentation live beside SQL. Agents keep models thin and tested, prefer incremental strategies for large facts, and never hardcode environment-specific database names inside model bodies.

sources.yml -> staging models -> intermediate -> marts
                     |
              dbt compile/run/test
                     |
              warehouse relations + docs

When to use this skill

• Versioned SQL transforms in Snowflake/BigQuery/Redshift/DuckDB/etc.
• Data tests (unique, not_null, relationships, custom)
• Documenting marts for analysts