Combine Skills and MCP to Close the Context Gap — Pedro Rodrigues, Supabase

2026-05-15 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

Superbase has developed and released a new agent skill designed to guide AI agents, specifically Claude Sonnet 4.6, in correctly and safely interacting with the Superbase platform. This initiative addresses common agent shortcomings, such as missing security pitfalls like Row Level Security (RLS) in PostgreSQL, operating on stale training data, and failing to seek fresh information. The Superbase agent skill, announced today, was tested against a baseline and an MCP-only condition, demonstrating superior performance in task completeness scores across various models including Claude Code for Opus 4.6, Sonnet 4.6, GPT 5.4, and GPT 5.4 mini. The skill emphasizes principles like avoiding information duplication by pointing to up-to-date documentation, ensuring critical information is embedded directly in the `skill.md` file rather than easily skipped reference files, and being opinionated about optimal workflows for Superbase products.

Key takeaway

For AI Engineers integrating agents with complex platforms like Superbase, you should prioritize developing product-specific skills to provide explicit guidance. This approach ensures agents correctly handle critical aspects like security configurations (e.g., RLS) and follow optimized workflows, significantly outperforming agents relying solely on general training data or basic tool access. Implement robust testing with evaluation frameworks to validate skill effectiveness and iterate on your skill design.

Key insights

Agent skills provide essential guidance, improving AI agent performance and safety when interacting with complex products.

Principles

• Avoid duplicating existing documentation; point agents to the single source of truth.
• Embed critical, unchanging information directly in the skill's main file to prevent skipping.
• Be opinionated in skill design, guiding agents toward optimal product workflows.

Method

Superbase developed an agent skill, tested with Claude Sonnet 4.6, by comparing agent performance with and without the skill using a test completeness score across six scenarios and four agents from two vendors.

In practice

• Expose documentation via SSH to allow agents to navigate like a file system.
• Use evals to test agent behavior and documentation effectiveness.
• Start skill development minimally and iterate through versions.

Topics

• Superbase Agent Skill
• AI Agent Guidance
• Model Control Plane
• Row Level Security
• Evals (Agent Evaluation)

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• MCP vs Agent Skills: Different Altogether — MCP and Agent Skills are complementary, not competing, technologies for AI agent development.
• Evals Skills for Coding Agents — Improving infrastructure around AI agents, especially evaluation capabilities, is more critical than solely improving the underlying model.
• Soft Forks: How Agent Skills Create Specialized AI Without Training — Agent Skills enable specialized AI behavior at runtime through context injection, offering a cost-effective alternative to fine-tuning.
• AI 101: From Prompt Engineering to Skill Engineering — Skill engineering, distinct from prompt engineering, optimizes AI agent performance by creating reusable, testable, and maintainable capability packages.
• Anatomy of Agent SKILLS — Agent "skills" prevent context bloat by progressively disclosing information, improving efficiency and modularity.
• Agent Skills Masterclass with Nufar Gaspar — AI agent skills are portable, human-readable playbooks for agents, enabling consistent task execution and organizational standardization.

Open in AIssential →

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