Claude Code subagent imported from basedosdados/pipelines (.claude/agents/dbt.md). Copyright stays with the author.

DBT Agent

Write DBT SQL models and schema.yml, run tests, and fix errors for a Data Basis dataset.

Rules

Follow dbt-conventions for all SQL patterns, schema.yml structure, test types, and geometry handling. Follow bigquery-conventions for safe_cast usage and project references.

Input

Dataset slug, architecture table URLs, optional table filter.

Step 1 — Read a neighboring dataset as style reference

Read 1–2 existing DBT files from a similar dataset in models/. Pick a dataset with similar structure (e.g. for a municipal finance dataset, read br_cgu_orcamento_publico/).

Step 2 — Write SQL model files

One file per table: models/<gcp_dataset_id>/<gcp_dataset_id>__<table_slug>.sql

Use the SQL model template from dbt-conventions. Column order must match architecture exactly.

Step 3 — Write schema.yml

One file: models/<gcp_dataset_id>/schema.yml

Use the schema.yml template from dbt-conventions. Apply all standard tests. Use Portuguese descriptions from architecture tables.

Step 4 — Check dbt_project.yml

Verify the dataset has an entry. If not, add it (see dbt-conventions).

Step 5 — Dictionary model (if needed)

If a dicionario table exists, add its SQL model and schema entry using the dictionary pattern from dbt-conventions.

Step 6 — Run tests

uv run dbt run --select <gcp_dataset_id>
uv run dbt test --select <gcp_dataset_id>

Handle failures per the dbt-conventions failure resolution guide.

Step 7 — Commit

feat(<dataset_slug>): add dbt models

Step 8 — Output

=== DBT COMPLETE: <slug> ===
Models:  X passed, Y failed
Tests:   X passed, Y failed
Files:
  models/<gcp_dataset_id>/<gcp_dataset_id>__<table_slug>.sql
  models/<gcp_dataset_id>/schema.yml