Why use XML tags in subagent prompts?

Using XML tags instead of standard Markdown headings helps Claude parse the system prompt more efficiently, reduces token usage, and provides a clearer structure for the agent's role and constraints.

What is a Claude Code subagent?

A subagent is a specialized, isolated Claude instance with a focused role and specific tool access, designed to perform tasks autonomously and return a final output to the main conversation.

How do I invoke a specific subagent?

Claude can automatically select a subagent based on its description if it matches your current task, or you can explicitly command it, for example: 'Use the code-reviewer subagent to check my changes'.

Can subagents ask the user questions?

No, subagents operate as black boxes and cannot use tools requiring user interaction, such as AskUserQuestion. They must complete their task based on the initial instructions provided.

Where are subagent files stored?

Subagents are stored as Markdown files in `.claude/agents/` for project-specific agents or in `~/.claude/agents/` for global user-level agents.

Subagent Orchestrator

Name: Subagent Orchestrator
Author: marcusgoll

bymarcusgoll

•

生産性とワークフロー

Creates and manages specialized, autonomous subagents within Claude Code to delegate complex development tasks through isolated workflows.

概要

This skill provides a comprehensive framework for architecting, configuring, and orchestrating Claude Code subagents. It enables developers to break down complex workflows into isolated, specialized tasks—such as code reviews, research, or security audits—using the Task tool. By defining specific system prompts, tool permissions, and model selections, it ensures autonomous execution without user interruption, returning high-quality, focused results to the main conversation and improving overall development efficiency.

主な機能

Multi-agent workflow orchestration using the Task tool
Automatic and explicit subagent invocation logic based on task descriptions
Granular tool permissioning for secure, least-privilege agent execution
18 GitHub stars
Standardized XML-based system prompt templates for optimal token efficiency
Project-level and user-level subagent configuration management

ユースケース

Executing context-heavy generation tasks that operate independently from the main user conversation
Delegating deep-dive research and API documentation lookups to isolated background agents
Automating multi-step code quality, performance, and security reviews using specialized roles

概要

主な機能

Multi-agent workflow orchestration using the Task tool
Automatic and explicit subagent invocation logic based on task descriptions
Granular tool permissioning for secure, least-privilege agent execution
18 GitHub stars
Standardized XML-based system prompt templates for optimal token efficiency
Project-level and user-level subagent configuration management

ユースケース

Executing context-heavy generation tasks that operate independently from the main user conversation
Delegating deep-dive research and API documentation lookups to isolated background agents
Automating multi-step code quality, performance, and security reviews using specialized roles