performance-optimization

Table of Contents

• Overview
• Quick Start
• Token Budget Model
• Progressive Loading Patterns
• Optimization Workflow
• Quality Checks

Overview

Plugin ecosystems face a fundamental constraint: the skill description budget (2% of context window, ~16k characters at 200k tokens). Every skill loaded into context reduces the space available for actual work. This skill provides patterns to minimize token footprint while preserving full capability.

Core Principles

1. Metadata-first discovery - Claude scans ~100 tokens of frontmatter to decide relevance before loading full content
2. Progressive disclosure - Essential content loads immediately; advanced content lives in modules loaded on-demand
3. Token budgeting - Track and enforce per-skill token limits aligned with the ecosystem budget
4. Context-aware delivery - Load depth matches task complexity