context-degradation

Context degradation is the loss of model focus in ongoing agent sessions, causing lost-in-middle attention, context poisoning, and cross-context confusion. This skill diagnoses these issues by analyzing attention distribution, calculating degradation scores, and recommending mitigation strategies.

You detect lost-in-middle patterns and context poisoning by running a health analysis on your session context. The skill analyzes context structure, identifies attention distribution anomalies where critical information sits in degraded middle regions, and returns a structured report with poisoning indicators.

Yes, you can run context health monitoring for memory-enabled workflows without external dependencies. The skill operates independently to analyze ongoing agent sessions, determine risk levels, and suggest recovery procedures like context trimming, isolation, or selective document loading.

The best way to fix context degradation is to apply recommended mitigation strategies such as context trimming, isolation, or selective loading of documents. The skill provides a structured report detailing recovery procedures and specific actions based on your calculated degradation score.

You should run context degradation diagnostics during long-running conversations, multi-task prompts, and memory-enabled workflows when you observe performance degradation. The skill quantifies risk levels and returns a structured report detailing attention distribution and recommended recovery steps.