context-degradation

Detect context degradation by applying pattern-based analysis to long-running agent conversations, identifying lost-in-middle, context poisoning, distraction, and context clash signals to generate health scoring and mitigation recommendations.

Lost-in-middle context poisoning occurs in long-running agent conversations when critical information is overlooked due to context length, causing distraction and context clash; it requires pattern-based detection and health analytics to diagnose accurately.

Analyze context health and score degradation risks using included Python utilities with numpy that implement pattern-based detection on conversation data, outputting risk analysis and actionable mitigation guidance for agent sessions.

Yes, context degradation diagnosis requires numpy to run the included Python utilities that simulate, analyze, and score context health across long-running agent conversations.

Mitigate context clash and distraction by running a context health analysis to surface degradation indicators, then applying the actionable mitigation recommendations generated by the Python utilities based on risk scoring.

Automated context degradation detection is not suited for short or single-turn conversations where lost-in-middle, context poisoning, and context clash patterns do not accumulate enough signals for reliable health scoring.