Audit my migrations for rollback safety. List every migration in the project and check whether each has a working down/rollback path. For each migration: write or fix the down migration so it reverses the up migration; flag the irreversible ones explicitly (dropped columns, destructive transforms) and document what manual recovery would require for those; and check ordering hazards, like a rollback that drops a table other migrations later depend on. Then actually run the full cycle against a scratch database seeded with realistic data: migrate all the way up, then roll back one step at a time, confirming each down migration runs without errors and leaves the schema in the expected state. Don't rewrite the history of migrations that have already run in any environment; if an old migration is broken, write the fix as documentation or a corrective migration, not an edit. Don't fake the rollback of destructive changes by restoring nothing; call them irreversible instead. Deliver: rollback status per migration (working, fixed, irreversible), the corrected down migrations, and verification: the up-then-down cycle output on the scratch database showing a clean result.