How does the skill estimate token usage?

It uses specialized density ratios: Code is estimated at characters/4, Markdown at words x 1.3, and JSON at characters/3.3 to provide high-precision budget tracking.

What happens when I hit my token budget?

The system monitors usage thresholds: at 80% it issues a warning to monitor optimization, and at 95% it forces summarization or archival before continuing the task.

Are summaries regenerated every time?

No, summaries are cached in a local directory (.studio/.cache/summaries/) to avoid unnecessary computation costs and ensure consistency across sessions.

Can I prevent specific files or code from being summarized?

Yes, you can use the #no-summarize tag for critical patterns or the #preserve tag for entries referenced by active tasks to keep them in Tier 1 regardless of age.

How does the tiered context system work?

It categorizes content by age: Tier 1 (recent) keeps 100% detail, Tier 2 (30-90 days) uses LLM-generated summaries, and Tier 3 (>90 days) retains only metadata for archival indexing.

Context Management

Name: Context Management
Author: twofoldtech-dakota

bytwofoldtech-dakota

0•

生产力与工作流

Manages Claude's context window through tiered content organization, token budgeting, and intelligent summarization.

The Context Management skill is a precision-engineered orchestration tool designed to solve the problem of context bloat in AI-assisted development. It implements a rigorous three-tier architecture that intelligently transitions content from full detail to summaries and metadata indexes based on age and relevance. By establishing soft and hard token limits across specialized pools—including learnings, plans, and tool outputs—it ensures long-running sessions remain performant and stable. This skill is essential for developers working on complex, long-term projects where maintaining a coherent history of decisions and patterns is critical without overwhelming the LLM's memory limits.

主要功能

010 GitHub stars

02Automated LLM-powered summarization of historical developer learnings

03Specialized token density estimation for code, JSON, and text content

04Granular token budget allocation for distinct project domains

05Proactive context optimization workflow with threshold-based alerts

06Three-tier content system (Full, Summary, Index) based on entry age

使用场景

01Preserving long-term project knowledge without exceeding context limits

02Preventing session crashes in large-scale projects by managing token accumulation

03Optimizing AI performance by prioritizing recent code changes over historical data

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add twofoldtech-dakota/studio context-management

For use in Claude.ai and ChatGPT

Download Skill