Context engineering for IDEs: Agents.md & agent skills

Context engineering for integrated development environments (IDEs) involves equipping coding agents with necessary project context to improve their efficiency and accuracy in understanding and editing codebases. This approach often combines a repository-level AGENTS.md file with specific agent skills, enabling agents to understand architecture, conventions, and constraints earlier in their session, thus reducing the need for costly discovery processes. AGENTS.md provides a spatial awareness of the codebase, outlining project overviews, tech stacks, and operational boundaries to improve execution accuracy, efficiency, and safety. Agent skills, on the other hand, supply reusable expertise for specific tasks, offering a progressive disclosure model that loads relevant skills as needed. Through experimentation, it was demonstrated that AGENTS.md significantly reduces discovery costs and enhances adherence to conventions, providing the best balance of cost, speed, and reliability. Meanwhile, agent skills can enhance architectural quality but do not guarantee error prevention unless applied consistently at the right time. Therefore, a practical sequence involves starting with AGENTS.md for general context and supplementing with agent skills for specialized, repeatable workflows, ensuring a balance between comprehensive context and efficient execution.