context-degradation

Context degradation in agent systems occurs when large conversation contexts cause unexpected performance drops. It manifests as lost-in-middle, context poisoning, distraction, confusion, and clash patterns, which this Skill detects and diagnoses during long-running agent conversations.

To debug agent failures under heavy contextual load, use this Skill to run lightweight context health analysis and anomaly detection. It identifies specific degradation patterns and provides structured recovery workflows to restore agent performance.

Architectural patterns to mitigate context degradation include compaction, masking, partitioning, and isolation. This Skill provides guidance on applying these patterns to sustain context reliability and manage runtime monitoring of agent performance.

Yes, you can use this Skill for architecture design reviews focused on context management. It supports detecting degradation patterns and provides structured health checks to evaluate context reliability before production deployment.

Yes, numpy is a required dependency to run the context degradation diagnosis scripts. It provides the numerical computing foundation needed for the lightweight context health analysis and anomaly detection components.

Detection of lost-in-middle and context poisoning patterns applies primarily to long-running conversations with large contexts. The Skill focuses on identifying these specific degradation patterns and providing mitigation guidance rather than automated context repair.