Context Optimization

Maximize effective reasoning density in agent context windows by allocating strict token budgets, eliminating redundant context via semantic deduplication, and structuring prompt sequences to eliminate lost-in-the-middle attention degradation.

When to Use

• Constructing multi-turn agent execution loops with growing conversation histories
• Injecting multi-document retrieval-augmented generation (RAG) contexts into system prompts
• Compacting multi-agent message transcripts before dispatching downstream tasks
• Mitigating attention dilution and instruction forgetting in high-capacity LLMs
• Staying within hard token budget ceilings while preserving critical grounding facts

Core Rules