dbt

dbt Command Runner

Codex: Read ../_shared/codex-tools.md before following this skill. Apply its tool, root and dispatch mappings throughout.

Quick dbt operations.

Authentication

dbt typically needs Snowflake credentials available in the shell before any command runs. Most projects use one of:

• A .env file: never source it. Hand the variables to the one dbt command that needs them: python3 -B "<claudna-root>/scripts/env_from_file.py" .env SNOWFLAKE_ACCOUNT SNOWFLAKE_USER SNOWFLAKE_PASSWORD -- dbt <verb> ... (<claudna-root> per ../_shared/claudna-root.md). Use whichever names the project's profiles.yml reads with env_var().
• A project-specific auth helper that exports SNOWFLAKE_* vars: ask the user to run it in the shell the session starts from; do not source it yourself.
• Direct env vars exported in ~/.zshrc / ~/.bashrc

If the project's CLAUDE.md documents an auth helper, use that. Otherwise ask the user how their dbt profile expects to authenticate.

Quick Commands

Build a model:

dbt build --select model_name

Run with upstream deps:

dbt build --select +model_name

Test only:

dbt test --select model_name

Compile (no run):

dbt compile --select model_name

Full refresh (for incremental):

dbt run --select model_name --full-refresh

Instructions

When user asks about dbt:

1. Confirm they're in the dbt project directory
2. Confirm Snowflake auth is loaded (check the project's auth helper or .env); see "Authentication" above
3. Run the appropriate command
4. Report results and any test failures