context-degradation

Context degradation occurs when lost-in-middle phenomena, context poisoning, or attention confusion degrade LLM agent performance on long documents. This Skill diagnoses these specific attention pattern failures and token budget limitations.

Fix the lost in middle issue by applying architectural patterns for effective context management, such as summarizing or partitioning context. These mitigation strategies prevent attention pattern failures within limited token budgets.

Context poisoning is a degradation pattern where corrupted information disrupts LLM attention mechanisms, leading to agent confusion and performance degradation. Diagnosing this attention pattern helps determine if context partitioning is needed.

Diagnose context degradation by analyzing agent failures for patterns like lost-in-middle, poisoning, distraction, confusion, and clash. Understanding attention mechanisms and token budget limitations is required to identify these specific failure modes.

Effective context management architectural patterns include summarizing and partitioning context to prevent degradation. Implementing these strategies requires understanding attention mechanisms and token budget limitations to maintain agent performance.

Partition context to prevent LLM agent failures when diagnosing context degradation issues like lost-in-middle or context poisoning. This mitigation strategy effectively manages token budget limitations and restores attention pattern reliability.