database-and-prisma-review

Run `prisma migrate status` to detect drift, then scan the migration SQL for destructive verbs like DROP COLUMN, SET NOT NULL, or RENAME. Classify each statement as safe, unsafe, or destructive, and rewrite risky changes using the expand-contract pattern with batched backfills.

Replace loops calling findUnique per row with a single findMany using include or select to fetch relations. For unbounded queries, add take and cursor pagination instead of loading entire tables into memory.

prisma migrate dev creates and applies migrations in local development only, while migrate deploy applies existing migrations in CI/CD and production. Never run migrate dev or migrate reset against staging or production databases.

A non-nullable column without a default or backfill fails because existing rows have no value. Add the column as nullable, backfill data in batches, then enforce NOT NULL in a later migration to avoid table locks.

Use upsert instead of create so re-running the seed updates existing rows rather than failing on duplicates. Ensure every upsert targets a field backed by a @unique or @@unique constraint to prevent duplicate inserts under concurrency.

Skip it for tasks unrelated to database persistence, pure query performance tuning (use a query performance review instead), or non-Prisma databases where a Prisma-specific lens does not apply. It reviews code but never executes destructive commands.