Which models are supported by these benchmarks?

The skill includes specific degradation thresholds and behavior patterns for Claude 4.5, GPT-5.2, and Gemini 3 series, informing better model selection for specific tasks.

Can this skill prevent AI hallucinations in long chats?

Yes, by applying patterns like isolation, relevance filtering, and explicit sectioning, it reduces the 'distraction' and 'confusion' that often lead to fabrications.

What is the 'lost-in-middle' phenomenon?

It is a U-shaped attention curve where models recall information at the beginning and end of a context window much better than information buried in the center.

How does this skill help with context poisoning?

It identifies when errors or hallucinations enter the context and provides strategies to truncate or verify information to stop compounding failures and feedback loops.

Context Degradation Diagnostics

Name: Context Degradation Diagnostics
Author: monmacllcapp

bymonmacllcapp

0•

Developer Tools

Diagnoses and mitigates performance failures in long-context AI interactions through proven architectural patterns and attention management.

This skill provides a comprehensive framework for identifying and resolving the predictable ways language models fail as context length increases. It equips developers with patterns to combat the "lost-in-middle" phenomenon, context poisoning, and informational distraction, ensuring agentic workflows remain reliable even during extended conversations. By leveraging empirical benchmarks and model-specific behavior analysis, the skill helps optimize information placement and architectural isolation to maintain high-quality reasoning and prevent hallucinations in complex, data-heavy sessions.

Key Features

01Implements architectural strategies like masking and partitioning to reduce distraction.

02Identifies 'lost-in-middle' attention gaps to optimize critical information placement.

03Analyzes context clash to resolve contradictory data from multi-source retrieval.

040 GitHub stars

05Provides model-specific degradation thresholds for Claude, GPT, and Gemini series.

06Detects and remediates context poisoning and compounding feedback loops.

Use Cases

01Designing resilient context management systems for production-grade AI agents.

02Evaluating the impact of large document ingestion on model recall and reasoning accuracy.

03Debugging agent performance drops during lengthy coding or research sessions.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add monmacllcapp/skill-forks context-degradation

For use in Claude.ai and ChatGPT

Download Skill