What is context compaction?

Context compaction is the practice of summarizing previous messages and tool outputs when approaching context limits to preserve essential information while freeing up tokens for new reasoning.

How does observation masking improve performance?

It replaces large, verbose tool outputs with compact summaries or references once they have served their purpose, preventing context bloat without losing track of the agent's progress.

Does this skill reduce LLM API costs?

Yes. By optimizing token usage through summarization and intelligent caching, it reduces the total number of tokens processed in each request, directly lowering operational costs.

When should I use the Context Optimization skill?

You should use it when context limits constrain task complexity, when you notice latency increasing in long conversations, or when building production-grade agents that need to handle large-scale data efficiently.

Context Optimization

Name: Context Optimization
Author: pCruvinel

bypCruvinel

0•

データサイエンスとML

Optimizes LLM context windows through strategic compaction, masking, and caching to improve model performance and reduce costs.

This skill empowers Claude to manage limited context windows more effectively by applying advanced context engineering techniques. It implements strategies such as compaction (summarizing history), observation masking (eliding verbose tool outputs), and KV-cache optimization to maximize the effective capacity of the model. This is essential for building production-scale AI agents, managing long-running conversations, and processing massive datasets where token efficiency directly impacts latency, reasoning quality, and operational expenses.

主な機能

01Context Compaction for high-fidelity history summarization

02Observation Masking to elide verbose tool outputs

03KV-Cache Optimization for improved prefix matching and lower latency

04Sub-agent Context Partitioning for complex task isolation

05Token Budget Management and trigger-based optimization

060 GitHub stars

ユースケース

01Maintaining reasoning quality when processing documents that exceed standard context limits

02Scaling long-running autonomous agents that accumulate large conversation histories

03Reducing API costs and latency in high-volume production LLM applications

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add pcruvinel/antig context-optimization

For use in Claude.ai and ChatGPT

Download Skill