context-engineering

To provide project-specific context, you define rules files for tech stack and conventions, then load relevant specification and architecture documents to guide agent output quality.

Poor agent output quality often stems from inadequate context. Curating structured context with rules files and relevant source files helps maintain focus and enforces project conventions.

You prevent context drift by managing long conversations actively, feeding error outputs and test results back to the agent for iterative improvement while maintaining relevant context.

Yes, you can load relevant sections of specification and architecture documents to provide targeted context, ensuring the agent adheres to defined specifications during code-based tasks.

You need rules files defining your tech stack, commands, conventions, and boundaries, alongside relevant documentation and source files to parse and apply project-specific context.

Feeding error outputs and test results to the agent provides direct feedback for iterative improvement, enhancing output quality by refining context based on actual execution failures.