@ericokuma can you do PROD check here, since you know exact steps to replciate

Let's try to cherry pick this into 0.78 if possible

FWIW, I see this error message in my terminal when connecting to MotherDuck:

2025-12-19T10:45:01.707 ERROR failed to migrate db {"error": "database/sql/driver: could not connect to database: Connection Error: Can't open a connection to same database file with a different configuration than existing connections"}

A few observations while reviewing this change:

Rollback coverage for connector file creation failure

If runtimeServicePutFile for the connector (lines 396-407) throws an error after .env has already been written and reconciled, the .env changes won't be rolled back. The current catch block (lines 466-472) only handles abort signals. This could leave orphaned secrets in .env if the connector file creation fails for other reasons (e.g., network error, permission issue).

A single try-catch covering both the .env write and connector file creation, with rollback logic in one place, would handle this more cleanly.

Inconsistent ordering between "Test and Connect" and "Save Anyway"

This PR fixes the ordering for "Test and Connect" (.env first, then connector), but saveConnectorAnyway still writes the connector file before .env. I understand "Save Anyway" bypasses reconciliation so the order doesn't matter today, but having consistent ordering across both paths would be easier to reason about. Not suggesting we expand scope here, but worth considering as a follow-up.

Test coverage

The checklist indicates this isn't covered by tests. Given the subtle ordering requirements here (secrets must exist before connector reconciliation), an integration test that exercises the welcome flow for MotherDuck would help prevent regressions.

Developed in collaboration with Claude Code

A few observations while reviewing this change:

Rollback coverage for connector file creation failure

If runtimeServicePutFile for the connector (lines 396-407) throws an error after .env has already been written and reconciled, the .env changes won't be rolled back. The current catch block (lines 466-472) only handles abort signals. This could leave orphaned secrets in .env if the connector file creation fails for other reasons (e.g., network error, permission issue).

A single try-catch covering both the .env write and connector file creation, with rollback logic in one place, would handle this more cleanly.

Inconsistent ordering between "Test and Connect" and "Save Anyway"

This PR fixes the ordering for "Test and Connect" (.env first, then connector), but saveConnectorAnyway still writes the connector file before .env. I understand "Save Anyway" bypasses reconciliation so the order doesn't matter today, but having consistent ordering across both paths would be easier to reason about. Not suggesting we expand scope here, but worth considering as a follow-up.

Test coverage

The checklist indicates this isn't covered by tests. Given the subtle ordering requirements here (secrets must exist before connector reconciliation), an integration test that exercises the welcome flow for MotherDuck would help prevent regressions.

Developed in collaboration with Claude Code

Addressed them!