Agentic Engineering: Skills vs. Agent.md Explained

Summary

Unlock the secrets to effective agentic engineering by understanding the distinct roles of "skills" and "agents.md" files. Skills operate on a need-to-know basis, with only their name and description loaded into the AI's context to save valuable tokens. This is ideal for specific, on-demand actions like "code structure" for refactoring code into a service layer architecture, or "code simplifier" to make complex code more readable. The "GP loop" skill is a powerful tool for iterative code review, analyzing feedback from tools like Greyle, suggesting fixes, and looping until a perfect 5 out of 5 score is achieved. In contrast, "agents.md" files are loaded into the AI's context at every turn, providing persistent information. This is used for crucial, always-on directives like establishing a specific branching and PR workflow or providing access to downloaded source code for up-to-date information, enabling the AI to search and reference it effectively. Both skills and agents.md are essential components for building sophisticated AI agents.