context-engineering

The context-engineering skill is designed to optimize how AI agents receive and process information to maximize their output quality. It addresses the common problem of agents either hallucinating responses due to insufficient context or losing focus when overwhelmed by irrelevant details. By structuring context deliberately, this skill ensures that agents see the right information at the right time, improving reliability and adherence to project conventions. It focuses on both the selection of context and the timing of its presentation, helping maintain consistency across coding sessions and development tasks.

The main features of this skill revolve around a hierarchical approach to context management. It supports persistent project-wide rules files, selective loading of specifications or architectural documents per feature, pre-task loading of relevant source code and tests, and iterative updates with error outputs and test results. This hierarchy allows agents to prioritize critical, trusted information while still considering transient data like recent conversation history or task-specific outputs. Trust levels for loaded files are also emphasized, distinguishing between trusted source code, files requiring verification, and untrusted external content. By doing so, it helps prevent agents from acting on potentially misleading or unsafe information.

This skill is particularly useful for software developers, AI-assisted development teams, and project managers who need their agents to follow coding standards, architectural patterns, and project-specific conventions. Use cases include starting a new coding session, troubleshooting declining agent performance, switching between codebase modules, and setting up new projects with clear rules and patterns. The skill is valuable whenever structured, context-aware agent behavior is needed to enhance productivity, maintain consistency, and reduce errors in automated development workflows.