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

The text discusses a skill designed to aid agents in using Supabase effectively by focusing on documentation access, security, tooling workflow, and schema management. It emphasizes the importance of checking current documentation before implementation, as Supabase's configurations and APIs frequently change. The skill includes a built-in checklist for security requirements and discourages using the Supabase MCP server with production databases, recommending local or staging databases instead. It highlights that agents often rely on outdated training data, underscoring the necessity of accessing up-to-date documentation and providing tools like the MCP server and CLI for schema management. Testing showed that using the skill improved performance across different agent models, indicating that the bottleneck is often context rather than capability. The skill is designed to be maintainable by teaching agents how to find information rather than including the information itself, and it encourages an opinionated approach to schema management to avoid unnecessary migrations. The document concludes with insights gained from testing and suggestions for developing similar skills, such as starting simple and expanding based on evaluation results and community feedback.