Agent Skills explained: Replacing AGENTS.md with reusable AI knowledge
Article Summary
Antoine van der Lee reveals why he ditched AGENTS.md files across all his projects. The replacement? A reusable AI knowledge system that's already supported by Cursor, Claude, and Gemini.
Van der Lee, creator of a 70+ lesson Swift Concurrency course, explains Agent Skills: an open format for packaging domain expertise that AI coding agents can reference across projects. Instead of maintaining duplicate AGENTS.md files in every codebase, developers can now install specialized skills once and reuse them everywhere.
Key Takeaways
- Agent Skills separate project-specific from reusable domain knowledge across codebases
- Already supported by Cursor, Claude, Gemini, and Codex CLI tools
- Skills use decision trees to guide agents through complex topics
- Deep domain expertise matters: skills need more than surface-level guidance
- Install via CLI, sync to projects, agents auto-reference when relevant
Agent Skills solve the sync problem of maintaining AI coding instructions across multiple projects by creating shareable, domain-specific knowledge packages that major AI tools already support.
About This Article
Antoine van der Lee had to keep separate AGENTS.md files in each of his projects. This meant manually updating coding standards, Swift preferences, and refactoring knowledge across every codebase.
Van der Lee built a reusable Agent Skill for Swift Concurrency by taking 70+ lessons from his course and organizing them into structured reference files with decision trees. Developers can install it through the openskills CLI and use it with Cursor, Claude, and Codex.
The Swift Concurrency Skill lets van der Lee automatically find and remove unnecessary @MainActor attributes by checking project settings like Default Actor Isolation and Swift version. Developers who used the skill reported positive results.
