Context Degradation

muratcankoylan/agent-skills-for-context-engineering

When your agent starts hallucinating facts it just read or ignoring critical instructions buried in a long prompt, you're hitting context degradation. This skill walks through five failure patterns: lost-in-middle (that U-curve attention problem where middle content gets ignored), poisoning (bad tool outputs contaminating everything downstream), distraction (irrelevant docs tanking performance even when correct info is present), confusion (mixing up constraints from different tasks), and clash (contradictory but valid sources). It's built around the insight that these aren't random failures but predictable engineering problems with specific detection signals. The placement strategies and filtering rules are directly applicable if you're building anything that needs to handle contexts over 4K tokens reliably.

skills.sh installs: 280
GitHub stars: 18.3K
last updated: May 26, 2026

View on GitHub

Install

$ npx -y skills add muratcankoylan/agent-skills-for-context-engineering --skill context-degradation --agent claude-code

Installs into .claude/skills of the current project.

Files

Context Degradation Patterns

Diagnose and fix context failures before they cascade. Context degradation is not binary — it is a continuum that manifests through five distinct, predictable patterns: lost-in-middle, poisoning, distraction, confusion, and clash. Each pattern has specific detection signals and mitigation strategies. Treat degradation as an engineering problem with measurable thresholds, not an unpredictable failure mode.