context-engineering

Optimize AI agent context by applying a hierarchical strategy that loads rules files, specs, source files, and error outputs from most persistent to transient. This structure prevents information overload and hallucinations by ensuring agents receive the exact context needed for peak performance.

Provide project context to AI agents during task switches by using tailored context packing strategies like brain dumps or selective includes. This approach quickly re-establishes the correct context, maintaining output quality across different project modules.

AI agents lose focus and hallucinate during long sessions because they receive too much transient information without hierarchical organization. Recover degraded agent performance by reloading conversation history and source files using a persistent-to-transient context loading strategy.

Manage ambiguities between specs and existing implementations by applying confusion management strategies that detect and surface mismatches. This approach prompts you for clarification, ensuring the AI agent operates with an accurate understanding of the codebase rules.

Reinitialize context rules for an AI-assisted development session when starting new projects, recovering from declining performance, or switching between project modules. Establishing project-wide rules early ensures consistent context packing and optimized agent behavior across tasks.