context-degradation

Detect lost-in-middle attention degradation by analyzing multi-turn conversations to identify where critical information is dropped from context windows, providing a context health score and actionable recommendations to improve AI robustness.

Context poisoning is the corruption of reliable context by misleading inputs in long-running sessions. Diagnose it by analyzing context structure and poisoning indicators to calculate a health score and provide mitigation recommendations.

Yes, you can monitor context health across multi-turn conversations using this skill with only numpy installed, as it applies detection algorithms to long-running agent sessions to identify degraded attention and distraction.

Your language model loses critical information due to context degradation, where attention degrades over long sessions. Analyzing context structure and applying context health scoring identifies lost-in-middle patterns to prevent unreliable outputs.

Improve AI robustness against context degradation by running context health analysis to detect degraded attention and poisoning indicators, generating actionable recommendations to preserve critical information and maintain reliability across sessions.

Limitations in diagnosing context degradation include relying solely on numpy for numerical analysis and detecting patterns like lost-in-middle and distraction, which may not capture all nuanced context poisoning scenarios in complex agent sessions.