Skill Issue: How We Used AI to Make Agents Actually Good at Supabase — Pedro Rodrigues, Supabase

2026-05-04 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, extended

Summary

Pedro from Superbase presented a workshop on developing and testing agent skills, emphasizing their role in enhancing agent performance within products like Superbase. The workshop, titled "Level Up Your Skills," detailed the structure of skills, which are essentially folders containing `skill.md` markdown files with front matter for identification and description, and can reference other markdown or script files. A key concept discussed was progressive disclosure, where agents load only necessary information to conserve context. The presentation highlighted the distinction between skills and MCP tools, noting that skills provide context and workflows, while MCP tools handle integrations and run in a server-side environment. Pedro demonstrated building a Superbase skill to address a Role-Level Security (RLS) issue in a performance review application, specifically ensuring a `security invoker` flag is used when creating PostgreSQL views to enforce RLS policies. The workshop also covered automated testing of skills using evaluations (evals), following an OpenAI-proposed framework for defining metrics, creating skills, running tests, and grading performance in an iterative cycle.

Key takeaway

For AI Engineers developing agent-friendly applications, understanding and implementing agent skills is critical for managing context and defining complex workflows. You should leverage skills to provide agents with specific, progressively disclosed information, ensuring robust behavior, especially when dealing with database security like PostgreSQL RLS. Automate skill testing using evaluation frameworks to maintain reliability and ensure consistent agent performance across different environments and updates.

Key insights

Agent skills enhance LLM performance by providing context and workflows through progressive disclosure.

Principles

• Progressive disclosure optimizes agent context management.
• Skills define workflows; MCP tools handle integrations.
• Automated evaluations are crucial for skill reliability.

Method

Develop skills using `skill.md` with front matter and reference files. Test with an eval-driven framework: define metrics, create skill, run evaluations (manual/automated), and iterate based on grading.

In practice

• Use `security invoker` flag for PostgreSQL views with RLS.
• Employ "use [skill name]" in prompts to ensure skill loading.
• Implement evals with input, expected output, and assertions.

Topics

• Supabase
• AI Agents
• Agent Skills
• Progressive Disclosure
• Role-Level Security

Best for: AI Engineer, MLOps Engineer, Software Engineer

Related on AIssential

• openai / skills — Agent Skills standardize AI agent capabilities for repeatable task execution across various applications.
• My approach to agentic skills — Effective agentic skill management involves careful design, context optimization, and a feedback loop for self-correction.
• addyosmani / agent-skills — AI coding agents can achieve production-grade software quality by following structured, verifiable engineering workflows and best practices.
• Combine Skills and MCP to Close the Context Gap — Pedro Rodrigues, Supabase — Agent skills provide essential guidance, improving AI agent performance and safety when interacting with complex products.
• DAY 3 Livestream - 5-Days of AI Agents: Intensive Vibe Coding Course With Google — Progressive disclosure of agent skills prevents context rot by loading minimal metadata and fetching full instructions only when needed.
• What AI Agent Skills Are and How They Work — AI agent skills provide procedural knowledge to LLMs via an open standard, enabling complex task execution and efficient resource loading.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.