What is context compression in Claude Code?

Context compression is a strategy used to summarize and prune session history to fit within token limits while retaining the most essential information for the AI to continue its task.

When should I trigger context compression?

The most effective trigger is usually at 70-80% context utilization, balancing the need for raw history with the necessity of staying within model limits.

Can I use this for very large codebases?

Yes, it enables a three-phase workflow—Research, Planning, and Implementation—specifically designed to handle projects that generate millions of tokens.

Why is 'Anchored Iterative Summarization' the recommended method?

It maintains structured sections and only summarizes newly-truncated content, merging it with existing summaries to preserve critical details more effectively than total regeneration.

How does this skill handle modified files during compression?

It uses an artifact trail solution that explicitly tracks created, modified, and read files to ensure the AI doesn't lose track of the codebase state during the summarization process.

Context Compression & Token Optimizer

Name: Context Compression & Token Optimizer
Author: eyadsibai

byeyadsibai

生産性とワークフロー

Optimizes long-running Claude sessions by implementing advanced context compression and structured summarization strategies to reduce token usage while preserving critical project state.

概要

Context Compression provides a robust framework for managing high-token Claude Code sessions, preventing context overflow while maintaining session intelligence. It offers multiple compression strategies—including Anchored Iterative and Regenerative summarization—to ensure that session intent, architectural decisions, and file modifications remain accessible even as the raw conversation history is pruned. This skill is essential for developers working on large-scale codebases where sessions can span millions of tokens, allowing for efficient multi-phase workflows from research to implementation without losing the artifact trail.

主な機能

0 GitHub stars
Structured summary formats tracking session intent, file changes, and decisions
Probe-based evaluation methods to test factual retention and planning continuity
Anchored Iterative Summarization for high-fidelity preservation of session state
Automated compression triggers based on 70-80% context utilization thresholds
Artifact trail management to prevent loss of file-state tracking in coding tasks

ユースケース

Maintaining a persistent memory of complex architectural decisions across long development cycles
Managing massive Claude sessions during deep exploration of enterprise-scale codebases
Reducing API costs by optimizing tokens-per-task through efficient summarization

概要

主な機能

0 GitHub stars
Structured summary formats tracking session intent, file changes, and decisions
Probe-based evaluation methods to test factual retention and planning continuity
Anchored Iterative Summarization for high-fidelity preservation of session state
Automated compression triggers based on 70-80% context utilization thresholds
Artifact trail management to prevent loss of file-state tracking in coding tasks

ユースケース

Maintaining a persistent memory of complex architectural decisions across long development cycles
Managing massive Claude sessions during deep exploration of enterprise-scale codebases
Reducing API costs by optimizing tokens-per-task through efficient summarization