What problem does it solve? AI agents degrade unpredictably as conversation context grows, producing wrong answers, hallucinations, and tool misuse that are hard to diagnose. This Skill provides a systematic framework for recognizing, measuring, and mitigating the five core degradation patterns: lost-in-middle, context poisoning, distraction, confusion, and clash. ## Core Features & Use Cases - Degradation Pattern Diagnosis: Identify lost-in-middle effects, context poisoning, distraction, confusion, and clash using documented symptoms and empirical benchmarks. - Context Health Scoring: Run the included Python analyzer to compute a composite health score from token utilization, attention distribution, and poisoning indicators. - Mitigation Architecture Guidance: Apply the Write/Select/Compress/Isolate strategies with concrete patterns like compaction, observation masking, and sub-agent isolation. - Use Case: When a long-running support agent starts giving contradictory answers after 50 turns, use this Skill to detect context poisoning, score context health, and apply truncation or compaction before quality collapses. ## Quick Start Ask the agent to analyze the current conversation context for degradation risks and recommend mitigation strategies using the context-degradation skill.
