Is iterative summarization better than full regeneration?

Yes, for long sessions. Iterative summarization only processes the newest content and merges it into an existing structured summary, which is faster and less likely to degrade technical accuracy over many compression cycles.

How does structured summarization prevent information loss?

By using specific sections for 'Files Modified', 'Decisions Made', and 'Next Steps', the summarizer is forced to populate a checklist. This prevents the silent drift where an AI might otherwise forget specific file paths or technical constraints.

When should I trigger context compression in a coding session?

The recommended strategy is to trigger compression when you reach 70-80% context utilization. This provides enough buffer to generate a high-quality summary without hitting hard model limits.

What is the primary benefit of tokens-per-task optimization?

Traditional compression targets tokens-per-request, but if critical info is lost, the agent must re-fetch it. Tokens-per-task optimization minimizes the total tokens consumed from start to finish by ensuring the agent never loses the data it needs to complete a job.

Context Compression Optimizer

Name: Context Compression Optimizer
Author: CoralShades

byCoralShades

0•

생산성 및 워크플로

Optimizes long-running AI agent sessions by implementing structured context compression strategies that preserve critical technical details and decision history.

The Context Compression skill provides advanced strategies for managing massive conversation histories and large codebases that exceed standard LLM context windows. By shifting the focus from simple token reduction to 'tokens-per-task' optimization, it ensures that essential project state—such as file modifications, architectural decisions, and error logs—remains accessible. It utilizes anchored iterative summarization and structured templates to prevent agents from losing their 'artifact trail,' ensuring seamless continuity and reducing the need for costly information re-fetching in complex, multi-step engineering workflows.

주요 기능

01Tokens-per-task optimization to maximize long-term efficiency and quality

02Anchored Iterative Summarization for persistent, structured context updates

03Artifact trail management to prevent loss of file-state information

040 GitHub stars

05Probe-based evaluation to measure functional retention of technical details

06Structured summary templates for tracking file changes, decisions, and intent

사용 사례

01Debugging complex issues where the agent must track changes across multiple files over time

02Developing and navigating large-scale codebases with 5M+ tokens

03Managing long-running agent sessions that exceed million-token history limits

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add coralshades/acm-ai context-compression

For use in Claude.ai and ChatGPT

주요 기능

01Tokens-per-task optimization to maximize long-term efficiency and quality

02Anchored Iterative Summarization for persistent, structured context updates

03Artifact trail management to prevent loss of file-state information

040 GitHub stars

05Probe-based evaluation to measure functional retention of technical details

06Structured summary templates for tracking file changes, decisions, and intent

사용 사례

01Debugging complex issues where the agent must track changes across multiple files over time

02Developing and navigating large-scale codebases with 5M+ tokens

03Managing long-running agent sessions that exceed million-token history limits