// the find
supabase/agent-skills
Agent Skills to help developers using AI agents with Supabase
A collection of context files ("skills") that teach AI coding agents how to work correctly with Supabase — covering auth, RLS, schema design, connection pooling, and Postgres query patterns. The install command drops the right files into whatever agent's context directory it recognizes. Aimed at developers who are tired of their AI assistant confidently writing broken Supabase auth or missing RLS on new tables.
The Postgres best-practices skill is genuinely well-structured — 8 categories with explicit priority levels (Critical/High/Medium), so an agent knows which guidance to weight more heavily. The `.well-known` discovery mechanism means Supabase's own tooling can serve these skills from their domain, which is a real distribution win over just 'put this file in your project'. The agent-skills open standard means it's not locked to one tool — the same files work in Claude Code, Cursor, Cline, and Copilot without conversion. Two focused skills rather than one giant blob keeps context window cost reasonable.
The 'supabase' skill is described as covering all of Supabase, but the actual references directory has almost nothing in it — just a feedback issue template and a skill-feedback file. All the real content is in the SKILL.md frontmatter, which means agents are getting high-level guidance without the detailed reference material that the postgres skill has. The agent-skills standard itself is barely six months old and the `.well-known` spec they reference is still an open PR (#254), so anything relying on the ecosystem tooling is betting on a standard that hasn't landed yet. No tests for whether the skills actually improve agent output — the test suite is a sanity check that the build artifacts are well-formed JSON, not a signal that the guidance is correct or that agents use it as intended.