AI Agents Know About Supabase. They Don't Always Use It Right.

Today we're releasing Supabase Agent Skills, an open-source set of instructions that teach AI coding agents how to build on Supabase correctly.

If you use Claude Code, Codex, GitHub Copilot, Cursor, or any of the agents that support Agent Skills Open Standard , you can install the skill in one command and get better results on every Supabase task.

AI agents already know a lot about Supabase.

We hit 100k stars on GitHub , and most models have seen plenty of Supabase code in their training data.

But knowing about Supabase and using it correctly are two different things.

We kept seeing agents skip RLS policies on exposed schemas, hallucinate CLI commands that don't exist (like supabase db execute ), create views without security_invoker = true (which silently bypasses RLS), and ignore the docs entirely, relying on training data that may be months out of date.

This is a follow-up to our earlier work on Postgres best practices for AI agents .

That release focused on general Postgres patterns.

Supabase Agent Skills goes further: it covers all Supabase products, encodes security requirements inline, and teaches agents to check the docs before implementing anything.