Drizzle Orm

Drizzle ORM (1.0 RC)

Use this skill for Drizzle 1.0 release candidates (drizzle-orm@rc, currently 1.0.0-rc.4), plus drizzle-kit, drizzle-seed, in-tree validators, Studio, and official Effect drivers.

Do not treat npm latest (0.45.x) as current for new work. Pin the rc channel for ORM + Kit + Seed together.

Workflow

1. Inspect the local Drizzle surface:
   • Exact versions / dist-tags for drizzle-orm, drizzle-kit, drizzle-seed, eslint-plugin-drizzle.
   • Dialect + driver entrypoints (postgres-js, neon-http, bun-sql, libsql, effect-postgres, …).
   • Schema files, defineRelations, drizzle.config.ts, migrations folder shape (v3 folders vs legacy journal).
   • Whether Effect (effect@beta + drizzle-orm/effect-*) or Promise drivers are in use.
2. For install, day-to-day usage, baselines, and troubleshooting, follow usage-guide.md first.
3. Refresh docs when versions drift or the task touches RQBv2, kit v3 migrations, codecs, or Effect. Start from source-map.md.
4. Route deeper detail:
   • Schema + SQL query builder: schema-queries.md.
   • RQBv2 defineRelations / db.query: relations-rqb.md.
   • Kit CLI, config, generate/migrate/push/pull/studio: kit-migrations.md.
   • Seed + Zod/Valibot/TypeBox/ArkType/Effect Schema: seed-validators.md.
   • Effect drivers vs @effect/sql-drizzle: effect-integration.md.
   • Driver matrix + 0.x → RC upgrade: drivers-upgrade.md.
5. Preserve the project's dialect and driver unless the user asks to switch.
6. Verify with the narrowest useful query, bunx drizzle-kit check / generate --explain-style dry runs, or a focused migrate against a disposable DB.

Core Judgment

• Always install @rc for new Drizzle 1.0 work: bun add drizzle-orm@rc and bun add -D drizzle-kit@rc. Never mix RC ORM with Kit 0.31.x.
• RQBv1 is gone. Use defineRelations(schema, …) and pass { relations } into drizzle() — not per-table relations() + { schema } for db.query.
• Prefer SQL query builder for joins/aggregations; prefer RQB for nested graphs.
• Casing is on tables (snakeCase.table / camelCase.table), not drizzle({ casing }).
• Production schema changes: generate → migrate. Use push only for local prototyping (--explain first).
• After leaving 0.x: run bunx drizzle-kit up once to convert migration folders to v3 (no journal.json).
• Validators live in-tree: drizzle-orm/zod, …/valibot, …/typebox, …/arktype, …/effect-schema. Avoid standalone drizzle-zod@0.x with RC.
• Effect on RC: use drizzle-orm/effect-* + Effect v4 beta (effect@beta, matching @effect/sql-*). Do not pair RC with @effect/sql-drizzle (Effect v3 / drizzle <0.50 only).
• Prefer $inferSelect / $inferInsert (aliases: InferSelectModel / InferInsertModel).

Verification

Prefer repository-owned commands. For meaningful Drizzle RC work, cover the relevant subset:

• Typecheck schema + relations + client factory.
• Focused insert/select/update/delete or db.query.*.findMany against a real or test DB.
• bunx drizzle-kit generate then review SQL; migrate on disposable DB; check for branch conflicts.
• Seed: reset + seed on a non-prod database only.
• Effect path: provide PgClient (or dialect) layer and run one Effect program end-to-end.
• After 0.x upgrades: confirm RQBv2 rewrite, casing builders, migrator folder v3, and codec-sensitive columns (arrays, timestamps, JSON).

Report which checks ran, which did not, and any @rc vs latest assumptions that remain.