supabase-postgres-best-practices
Review Postgres schema, queries, indexes and migrations using Supabase guidance.
npx skills add supabase/agent-skills --skill supabase-postgres-best-practicesConnect applications to databases, shape schemas and improve query performance.
Choose the skill for your database engine or ORM. Work on a branch or staging database when changing schema.
Review Postgres schema, queries, indexes and migrations using Supabase guidance.
npx skills add supabase/agent-skills --skill supabase-postgres-best-practicesUse Prisma Client for typed queries, relations and transactions.
npx skills add prisma/skills --skill prisma-client-apiConfigure Prisma for the application's database provider and connection.
npx skills add prisma/skills --skill prisma-database-setupConnect to and work with a Neon Postgres database, including development branches.
npx skills add neondatabase/agent-skills --skill neon-postgresSet up a Convex backend and wire queries and mutations into an application.
npx skills add convex/agent-skills --skill convex-quickstartConnect an application to Supabase and work with its database, authentication and storage features.
npx skills add supabase/agent-skills --skill supabaseNo skill in this category has a full write-up yet.
A database skill gives an agent two things it otherwise improvises: how to look before it writes, and how to undo. Reading a schema is easy. Knowing that a migration takes a lock, that a column rename breaks a dashboard nobody mentioned, or that staging seed data is not the shape of production — that is what the skill is for.
The work splits cleanly: query-and-explain tasks (read-only, useful immediately) and schema changes (need a plan, a backup, a rollback).
Pick by engine first, then by task.
Postgres on Supabase — supabase-postgres-best-practices for schema, index and query review; install supabase as well only if the app also uses its auth or storage. Postgres on Neon — neon-postgres, which knows about development branches; that is the feature worth having when you want to try a migration without touching the main database. Prisma projects — prisma-database-setup configures the provider and connection string, prisma-client-api is for typed queries, relations and transactions. A new project needs setup first; an existing one usually only needs the client skill. Convex — convex-quickstart is a backend platform with its own queries and mutations, so it is not interchangeable with the Postgres skills at all.
SELECT returning the same rows first, and a row count.LIMIT 1 and read the actual column names before trusting generated SQL.Cloud skills cover the deploy that usually triggers the migration. Coding skills covers the tests that should have caught the schema change before it shipped.
Review migrations, confirm rollback behaviour and compare representative query plans before applying changes.
Review the generated SQL, run it on a branch, test the rollback, and compare query plans before and after for the two or three queries the change targets.
Use the command beside the skill, select a supported agent, then reload the agent and confirm it can see the skill in your project.
Use supabase-postgres-best-practices for schema and query review on Supabase. Use neon-postgres for branch-aware connection work on Neon. If your application already uses Prisma, use prisma-client-api for queries and transactions. Add the supabase skill if your app also uses Supabase authentication or storage.
prisma-database-setup configures the provider and connection, prisma-client-api is for writing queries, relations and transactions. A new project needs setup first; an existing project usually only needs the client skill.
Not safely. Run the migration and its rollback on a branch or staging database first, time it there, and keep the generated SQL for review before it touches production.
No. Convex is a backend platform with its own queries and mutations. Install it for a Convex project only, and use the Postgres skills for Postgres databases.
Give it a read-only role on production and a writable one on staging. The failure you are protecting against is not malice — it is a plausible query with a missing clause.
Count first: SELECT COUNT(*) with the same WHERE clause as the UPDATE or DELETE, then compare with the affected-rows report afterwards.