What is Anchored Summarization in Claude Code?

It is a structured compression technique that forces the preservation of specific categories like session intent, file changes, and next steps to prevent critical information loss during long sessions.

How is the quality of compressed context validated?

The skill uses probe-based evaluation, testing the compressed context against functional queries to ensure a probe pass rate of over 90% before continuing.

When does the context compression trigger?

Compression is automatically triggered when the effective context window reaches 70% capacity, with a target of reducing utilization back to 50% while maintaining accuracy.

Does this skill lose recent conversation history?

No, the system is configured to always preserve at least the last 5 messages uncompressed to maintain immediate conversational continuity and flow.

Context Compression & Token Management

Name: Context Compression & Token Management
Author: yonatangross

byyonatangross

•

개발자 도구

Optimizes Claude Code sessions by condensing conversation history using structured, anchored summarization to prevent token limit degradation.

The Context Compression skill is a sophisticated memory management utility designed for long-running AI development sessions. It prevents 'context drift' and token overflow by implementing Anchored Iterative Summarization, which preserves essential session data—such as project intent, file modifications, and architectural decisions—while discarding redundant history. By utilizing probe-based validation and specific utilization triggers, this skill ensures that Claude maintains high performance and task accuracy during complex, multi-step engineering workflows without the degradation typically associated with long context windows.

주요 기능

0169 GitHub stars

02Functional probe-based validation to ensure zero information loss

03Anchored Iterative Summarization for structured data preservation

04Automated token utilization triggers at 70% capacity

05Effective context window calculation including system overhead

06Smart preservation of the most recent message history for continuity

사용 사례

01Managing multi-hour development sessions with extensive file modifications

02Persistent memory management for complex, multi-step debugging tasks

03Reducing latency and token costs in long-running agent workflows

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit context-compression

For use in Claude.ai and ChatGPT

Download Skill