When a dbt Cloud job fails and the error message isn't clear, this walks you through systematic diagnosis using the Admin API, run logs, and git history. The workflow is opinionated about one thing: never modify a test to make it pass without understanding why it failed. It covers three failure types (infrastructure, code, data) and includes specifics like how to construct artifact URLs when you don't have MCP access and when to use the discovering-data skill for test failures. The "Rationalizations That Mean STOP" table is pointed and useful, calling out the "board meeting in 2 hours" excuse and the "probably just flaky" handwave.
npx skills add https://github.com/dbt-labs/dbt-agent-skills --skill troubleshooting-dbt-job-errors