What are 'probes' in the context of this skill?

Probes are functional tests that verify if the compressed context still contains the necessary information to complete the current task, targeting a 90% accuracy rate.

Will I lose recent progress if I compress the context?

No. The skill is configured to always preserve the most recent messages uncompressed and uses incremental merging to ensure new progress is added to the summary rather than replacing it.

When is context compression automatically triggered?

Compression is typically triggered when you reach 70% of your effective context capacity, with a goal of reducing utilization to 50% while keeping the last 5 messages intact.

How does this skill differ from standard AI summarization?

Unlike generic summarization, this skill uses 'Anchored Summarization,' which forces the model to preserve specific, non-negotiable categories like session intent, decisions made, and pending blockers.

Context Compression

Name: Context Compression
Author: yonatangross

byyonatangross

•

생산성 및 워크플로

Optimizes Claude's memory by compressing conversation history while preserving critical task data and session intent.

The Context Compression skill is a sophisticated memory management tool for Claude Code, designed to handle long-running agent sessions that approach token limits. By utilizing Anchored Iterative Summarization, it structures history into key categories like session intent, file modifications, and architectural decisions, ensuring that essential information is never lost. This skill is vital for developers working on complex, multi-step workflows where maintaining task continuity and 'tokens-per-task' efficiency is more important than simple character reduction.

주요 기능

01Anchored Iterative Summarization for structured memory retention

02Incremental merging of session updates to prevent information degradation over time

03Automated compression triggers based on 70% context window utilization

04Continuity preservation for system prompts and the most recent message history

05Probe-based validation to ensure >90% information integrity post-compression

0629 GitHub stars

사용 사례

01Managing long-running agent sessions for large-scale code refactoring and migrations

02Maintaining architectural state across multi-step autonomous software development workflows

03Handling massive tool outputs and terminal logs without hitting token limits

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/skillforge-claude-plugin context-compression

For use in Claude.ai and ChatGPT

Download Skill