Prisma ORM Pro

A focused skill for building and operating Prisma in production — schema modeling, migrations, transactions, and the query patterns that hold up under serverless load and high request volume. It targets TypeScript backend engineers running Prisma 6 against Postgres (or MySQL) who keep hitting the same walls: connection exhaustion on Lambda, accidental N+1 queries, and migrations that work in dev but stall on deploy.

When to use it

• You're designing a schema.prisma with non-trivial relations (many-to-many, self-relations, composite keys) and want the @relation wiring correct the first time.
• Your serverless functions are throwing too many connections or Can't reach database server.
• A list endpoint is slow and you suspect a hidden per-row query.
• You need to choose between migrate dev and migrate deploy, or recover from a failed/drifted migration.
• You're adding interactive transactions and need correct isolation and timeout handling.

What it does

It works against the real Prisma surface area: the prisma CLI (migrate dev, migrate deploy, migrate resolve, db push, generate, studio), the schema.prisma file (datasource, generator, relationMode = "prisma" vs "foreignKeys"), and PrismaClient runtime APIs. On the query side it pushes you toward include/select and relation loading to kill N+1, batch transactions via prisma.$transaction([...]), and interactive transactions via prisma.$transaction(async (tx) => {...}, { timeout, isolationLevel }). For scale it covers connection pooling through the ?connection_limit= and ?pgbouncer=true params, the @prisma/adapter-* driver adapters, and Prisma Accelerate (@prisma/extension-accelerate) for connection pooling plus query-level cacheStrategy.

Example

curl -fsSL --create-dirs -o ~/.claude/skills/prisma-orm-pro/SKILL.md https://claudeskil.com/skills/prisma-orm-pro/SKILL.md

# dev: create + apply a migration from schema changes
npx prisma migrate dev --name add_orders

# prod: apply already-generated migrations, no prompts
npx prisma migrate deploy

// Accelerate + a cached, relation-loaded query (no N+1)
import { PrismaClient } from '@prisma/client'
import { withAccelerate } from '@prisma/extension-accelerate'

const prisma = new PrismaClient().$extends(withAccelerate())

const users = await prisma.user.findMany({
  include: { orders: true },              // single JOIN-style fetch
  cacheStrategy: { ttl: 60, swr: 30 },    // Accelerate edge cache
})

Tips & gotchas

• relationMode = "prisma" skips DB-level foreign keys (needed for PlanetScale) but means Prisma emulates referential actions and you must add @@index on FK columns yourself.
• Never run migrate dev in CI/prod — it can reset the database. Use migrate deploy.
• In serverless, instantiate one PrismaClient per container (module scope), not per request, and cap connection_limit low (often 1–5) behind a pooler.
• A failed migration blocks all future deploys until you run prisma migrate resolve --applied or --rolled-back.

FAQ

db push or migrate dev? db push syncs schema to DB with no migration history — good for prototyping. Use migrate dev once you need versioned, reviewable migrations.

How do I find an N+1? Set log: ['query'] on PrismaClient; repeated identical SELECTs per row mean you should fold the relation into one include/select.

Do I still need PgBouncer with Accelerate? No — Accelerate provides the pool. Use one or the other, and set pgbouncer=true when going through PgBouncer directly.