Deliver and operate

Publish, promote and apply a change, manage the data that travels with it, and diagnose a failed run.

Delivery work should leave an observable result: a sealed artifact, a recorded revision, an observation of the target, or a diagnostic tied to a stable error fragment. Choose the task that owns that result. Proving the change correct before it gets here is testing and safety.

Common tasks

Deliver a schema changeCarry one reviewed change from publish to a verified database.

Load controlled dataReconcile versioned lookup rows separately from ordinary schema changes.

Reconcile in KubernetesWhat the operator converges, and what it does not deploy.

Publish an OCI artifactPackage, verify, and distribute migration or schema artifacts through a registry.

Troubleshoot a failureStart from the exact error or symptom and verify the repair.

Choose the safety boundary

Use a throwaway database for tests, a dev or shadow database for replay and planning, and a reviewed target URL for a real apply. These roles are not interchangeable; database URLs and dev databases defines each one.

Commands that mutate a database keep their warning and verification step on the task page where the operation runs.

Reference

Use exit codes for automation contracts and test-case reference for the exact case format.