agent-stack-template / rules

---
description: Prisma schema and migration conventions
globs: prisma/**/*.prisma,prisma.config.ts
alwaysApply: false
---

# Prisma schema

1. **Tenancy.** Every business model carries `workspaceId`, a relation with
   `onDelete: Cascade`, and an index that starts with `workspaceId`. Without
   the cascade, deleting a workspace orphans rows; without the index, every
   scoped list is a sequential scan.
2. **Both sides of a relation** are declared, with explicit `@relation`.
3. **Ids**: `@id @default(cuid())` for business models. Better Auth models keep
   the ids the adapter assigns — do not change them.
4. **Timestamps**: `createdAt @default(now())` and `updatedAt @updatedAt` on
   anything a user edits.
5. **Indexes** on every field used in a `where`, `orderBy` or join. Compound
   indexes follow query order (`@@index([workspaceId, status])`).
6. **Uniqueness** with `@unique` / `@@unique`. Case-insensitive uniqueness has
   no direct equivalent — enforce it in the domain and say so in a comment.
7. **`@@map`** to snake_case table names; the Better Auth models must keep the
   names the adapter expects.

## Migrations

- `npm run db:migrate` in development; the generated SQL is reviewed, not
  trusted blindly.
- A rename that Prisma renders as drop + add is data loss. Split it: add the
  column, backfill, then remove the old one in a later migration.
- `npm run db:push` is for local scratch work only. It never runs against a
  shared database.