What is Anchored Iterative Summarization?

A strategy where only new conversation segments are summarized and then incrementally merged into a persistent, structured summary rather than regenerating the entire summary from scratch.

What is the difference between tokens-per-request and tokens-per-task?

Tokens-per-request measures single call efficiency, while tokens-per-task measures total tokens used to finish a goal, accounting for expensive context re-fetching costs caused by information loss.

When should I trigger context compression?

It is best to trigger compression when context utilization reaches 70-80% or at logical task boundaries to maintain a predictable context size without losing immediate session signals.

How does this skill prevent Claude from forgetting modified files?

It utilizes structured summarization with dedicated 'Files Modified' and 'Artifact Trail' sections, forcing the model to explicitly track and preserve file states during each compression cycle.

Context Compression & Token Optimization

Name: Context Compression & Token Optimization
Author: monmacllcapp

bymonmacllcapp

0•

生産性とワークフロー

Optimizes long-running Claude sessions by implementing structured context summarization and token-per-task management.

This skill provides advanced strategies for managing massive conversation histories and large codebases that exceed context window limits. By prioritizing 'tokens-per-task' over simple 'tokens-per-request,' it implements structured, anchored iterative summarization to preserve critical technical details like modified file paths, architectural decisions, and error states. It ensures Claude maintains high-fidelity continuity during complex, multi-step engineering tasks without losing the 'artifact trail' essential for successful code implementation and debugging.

主な機能

01Advanced compression trigger strategies including sliding window and task-boundary

02Probe-based evaluation framework to measure functional compression quality

030 GitHub stars

04Anchored Iterative Summarization to preserve session intent and file modifications

05Structured summary templates including state tracking and next steps

06Token-per-task optimization to minimize re-fetching costs and information loss

ユースケース

01Debugging complex system errors that require tracking changes across dozens of files

02Navigating and modifying codebases with over 5 million tokens of source material

03Managing long-running agent sessions that exceed the standard context window

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add monmacllcapp/skill-forks context-compression

For use in Claude.ai and ChatGPT

Download Skill