How does progressive disclosure save tokens?

It prevents Claude from reading every reference document at the start of a session, only loading specific files when a query requires that detailed information.

Where are the reference documents stored?

References are stored within the references/ subdirectory of the skill, allowing for a structured and organized documentation library.

Does this work automatically?

Claude is instructed to load these references only when the user's request specifically demands more detailed information than what is in the primary skill file.

Is this skill compatible with other Claude Code plugins?

Yes, this is a pattern that can be integrated into any complex Claude Code skill to improve its efficiency and focus.

Can I add more reference files to this skill?

Yes, you can expand the library by adding markdown files to the references folder and updating the SKILL.md to point to them.

On-Demand Reference Loader

Name: On-Demand Reference Loader
Author: yoshiwatanabe

byyoshiwatanabe

学习与文档

Optimizes context window usage by implementing a progressive disclosure pattern for loading detailed documentation only when required.

关于

The On-Demand Reference Loader skill is a performance-focused implementation pattern designed for Claude Code that allows developers to maintain extensive documentation libraries without overwhelming the AI's context window. By initially loading a lightweight entry point and fetching detailed markdown files from a reference directory only upon specific user request, this skill reduces token consumption, improves response speed, and ensures Claude remains focused on the primary task while keeping deep knowledge accessible for complex workflows.

主要功能

Token-efficient context management
Progressive disclosure pattern for document loading
0 GitHub stars
On-demand retrieval of detailed guidelines and references
Directory-based documentation structure
Optimized performance for complex skill sets

使用场景

Providing deep-dive documentation only when troubleshooting specific errors
Scaling AI agent capabilities across many reference files without context overflow
Managing large internal coding standards without cluttering every prompt

关于

主要功能

Token-efficient context management
Progressive disclosure pattern for document loading
0 GitHub stars
On-demand retrieval of detailed guidelines and references
Directory-based documentation structure
Optimized performance for complex skill sets

使用场景

Providing deep-dive documentation only when troubleshooting specific errors
Scaling AI agent capabilities across many reference files without context overflow
Managing large internal coding standards without cluttering every prompt