When should I use a subagent instead of a skill?

Subagents are recommended for tasks requiring parallel execution (more than 3 independent tasks), high security isolation, or long-running background processes.

Can I integrate MCP tools with this skill?

Yes, it includes specific workflows for wrapping MCP tools within skills so they are loaded on demand rather than bloating the primary context.

What is a Claude Code Skill?

A Claude Code Skill is a specialized capability that provides domain-specific guidance, best practices, and implementation patterns to enhance Claude's behavior during complex tasks.

What is the Three-Layer Architecture?

It is a design pattern that categorizes capabilities into MCP (External Connections), Skills (Task Specialization), and Subagents (Context Isolation) for optimal system organization.

How does progressive disclosure optimize AI performance?

It implements a tiered loading strategy—Index, Metadata, and Full Content—ensuring the agent only consumes the tokens it needs for the current task, reducing costs and improving speed.

Agent Skill Architect

Name: Agent Skill Architect
Author: ShunsukeHayashi

byShunsukeHayashi

•

生産性とワークフロー

Guides the creation, management, and optimization of AI agent skills using progressive disclosure and multi-agent design patterns.

Agent Skill Architect provides a standardized framework for building highly efficient AI agent capabilities within autonomous development environments. By implementing advanced patterns like three-layer architecture and progressive disclosure, it helps developers minimize token usage while maximizing functional power. This skill streamlines the entire lifecycle—from requirement definition and directory structuring to MCP tool wrapping and subagent delegation—ensuring that autonomous systems remain organized, scalable, and economically governed.

主な機能

01Context optimization via Progressive Disclosure to minimize token overhead

02Automated MCP tool wrapping for demand-based resource loading

03Standardized skill creation following the Three-Layer Architecture

04Comprehensive validation checklists for skill quality assurance

0513 GitHub stars

06Multi-agent system design and subagent delegation logic

ユースケース

01Designing complex multi-agent workflows for parallel task execution

02Building specialized domain capabilities for autonomous AI agents

03Reducing agent latency and costs by optimizing context window usage

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shunsukehayashi/miyabi agent-skill-use

For use in Claude.ai and ChatGPT

Download Skill