While testing agents on dbt projects, I wanted a better way to make sure their changes didn’t blow up the current model or something downstream.
There are tools that do this in CI, but I wanted something I could run locally before I even opened a PR. So I built this:
It works with an existing dbt Core + Snowflake project. It builds the changed models in temporary schemas, compares them against production, and cleans everything up afterward.
It checks:
- Row counts, null rates, cardinality, ranges, and percentiles
- Columns being removed, renamed, or changing types
- Added, removed, null, or duplicate primary keys
- Existing dbt test failures
- Downstream model and exposure impact, including column-level lineage
It runs locally and the repo is Apache 2.0.
Looking for feedback from people using agents with dbt Core + Snowflake. Anything missing? The use case here that I enjoy is a /goal prompt where the agent works towards the pipeline’s goals while fixing any issues raised from this cli.