Can sub-agents interact directly with the user?

No, sub-agents are designed to be autonomous. They cannot use tools like AskUserQuestion or wait for user input; they provide a final report back to the main chat or parent agent.

Where are these sub-agent configuration files stored?

Sub-agent files are stored in .claude/agents/ for project-specific tasks, ~/.claude/agents/ for user-wide tasks, or within plugin directories for shared utilities.

How do I choose the right model for a sub-agent?

Select 'sonnet' for reasoning and generation, 'haiku' for simple formatting or fast checks, and 'opus' for tasks requiring deep, complex logic and high context.

Can I run multiple sub-agents at the same time?

Yes, you can use the Parallel Agent pattern to launch several agents in a single message, allowing them to work on independent tasks simultaneously before collecting their outputs.

Claude Code Sub-Agent Architect

Name: Claude Code Sub-Agent Architect
Author: atournayre

byatournayre

0•

生产力与工作流

Guides the creation and configuration of specialized, high-performance sub-agents for autonomous multi-agent orchestration.

This skill provides a comprehensive framework for designing and deploying specialized sub-agents within the Claude Code ecosystem. It enforces critical constraints—such as ensuring agents operate autonomously without direct user interaction—and provides standardized templates for frontmatter, XML-based system prompts, and tool permissions. By leveraging this guide, developers can build complex workflows involving parallel execution and background tasks, enabling Claude to handle deep research, code generation, and specialized analysis more effectively while maintaining a clean main chat interface.

主要功能

010 GitHub stars

02Strict tool-set configuration following the principle of least privilege

03Structured file management for project-level and global agent personas

04Standardized XML prompt engineering for domain-specific sub-agent roles

05Implementation patterns for background and parallel agent execution

06Model selection optimization across Sonnet, Haiku, and Opus for cost and logic

使用场景

01Building an autonomous security auditor that scans code in the background

02Orchestrating parallel agents to analyze multiple API modules simultaneously

03Creating a dedicated documentation agent that generates READMEs from code analysis

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add atournayre/claude-personas make-subagent

For use in Claude.ai and ChatGPT

主要功能

010 GitHub stars

02Strict tool-set configuration following the principle of least privilege

03Structured file management for project-level and global agent personas

04Standardized XML prompt engineering for domain-specific sub-agent roles

05Implementation patterns for background and parallel agent execution

06Model selection optimization across Sonnet, Haiku, and Opus for cost and logic

使用场景

01Building an autonomous security auditor that scans code in the background

02Orchestrating parallel agents to analyze multiple API modules simultaneously

03Creating a dedicated documentation agent that generates READMEs from code analysis