context-optimization

Context optimization extends the effective capacity of limited context windows through strategic compression, masking, caching, and partitioning. It focuses on maximizing the quality of the context rather than simply increasing its size. Effective optimization can significantly improve performance and reduce costs by using tokens more efficiently. This skill provides techniques for compaction, observation masking, KV-cache optimization, and context partitioning.

It also covers budget management, practical guidance on when and how to apply different optimization strategies, and performance considerations for evaluating their effectiveness. The goal is to make better use of available capacity without requiring larger models or longer contexts.

The skill includes examples, guidelines, and integration points with other related skills, offering a comprehensive approach to optimizing context for AI agents.

What It Does

Applies compaction, masking, caching, and partitioning strategies to optimize context windows for AI agents, improving performance and reducing costs.

When To Use

Use this skill when context limits constrain task complexity, optimizing for cost reduction, reducing latency, implementing long-running agent systems, handling larger documents or conversations, or building production systems at scale.

Installation

Copy SKILL.md to your skills directory