Context Optimizer

概要

The Context Optimizer skill provides a unified interface for sophisticated context engineering, designed to keep Claude-based agents efficient and accurate during long-running tasks. It implements a comprehensive 'Four-Bucket Framework'—Write, Select, Compress, and Isolate—to prevent context degradation and the 'lost-in-the-middle' effect. By automating compression techniques like observation masking and summarization, managing KV-cache stability, and maintaining persistent session states via a structured directory, it ensures the agent maintains high performance and lower API costs even as context windows fill up.

主な機能

• Automated detection and mitigation of context degradation patterns
• KV-cache optimization for reduced latency and API cost efficiency
• 1 GitHub stars
• Structured session management using scratchpads and persistent checkpoints
• Progressive context compression including observation masking and compaction
• Objective recitation techniques to keep goals in the model's immediate attention

ユースケース

• Reducing API costs by optimizing token usage and message serialization
• Managing long-running coding sessions where the context window reaches its limit
• Maintaining agent focus and accuracy during complex, multi-phase software implementations