How do I control a subagent's permissions?

You define permission modes in the YAML frontmatter, such as 'acceptEdits' to skip approval for file changes or 'plan' to explore the codebase without making any modifications.

What is the main difference between a subagent and a skill?

Subagents run in an isolated context window and are explicitly triggered via the Task tool for autonomous work, whereas skills share context with the parent conversation and provide guidance automatically.

Which model should I choose for my subagent?

Use Haiku for simple, high-speed tasks like summarization, Sonnet for complex coding and logic, and Opus for mission-critical tasks requiring the highest reasoning capabilities.

Can a subagent use any tool available to Claude Code?

Yes, you can specify exactly which tools (Bash, Glob, Read, Edit, etc.) a subagent has access to, or use a wildcard to grant all permissions.

How does Claude decide which subagent to dispatch?

Claude uses the 'description' field in the subagent's YAML frontmatter to determine if the agent's capabilities match the requirements of the task being dispatched.

Subagent Creator

Name: Subagent Creator
Author: timequity

bytimequity

Productividad y Flujo de Trabajo

Architects and implements autonomous subagents for Claude Code to handle complex, discrete tasks with isolated context and specialized tools.

Acerca de

The Subagent Creator skill provides a comprehensive framework for building custom autonomous agents within the Claude Code environment. It guides developers through the process of defining subagent metadata, including specialized tool access, model selection (Haiku, Sonnet, or Opus), and permission modes. By standardizing the YAML frontmatter and system prompt structure, this skill helps users create reliable agents that can be dispatched via the Task tool to handle specific workflows like code reviews, data summarization, or security audits in a fresh, isolated context window.

Características Principales

Best practice guidelines for restricting tool access to maintain security and focus
Configurable permission modes including acceptEdits, bypassPermissions, and plan-only
0 GitHub stars
Comprehensive comparison between subagents and skills for optimal architectural decisions
Standardized YAML frontmatter templates for defining tools, models, and permissions
Structured prompt engineering framework covering role definition, process, and output constraints

Casos de Uso

Creating a lightweight 'Documentation Summarizer' using the Haiku model to optimize cost and speed
Building a dedicated 'Security Auditor' agent with read-only permissions and deep scanning prompts
Developing an autonomous 'Refactor Agent' that auto-accepts file edits for routine cleanup tasks

Acerca de

Características Principales

Best practice guidelines for restricting tool access to maintain security and focus
Configurable permission modes including acceptEdits, bypassPermissions, and plan-only
0 GitHub stars
Comprehensive comparison between subagents and skills for optimal architectural decisions
Standardized YAML frontmatter templates for defining tools, models, and permissions
Structured prompt engineering framework covering role definition, process, and output constraints

Casos de Uso

Creating a lightweight 'Documentation Summarizer' using the Haiku model to optimize cost and speed
Building a dedicated 'Security Auditor' agent with read-only permissions and deep scanning prompts
Developing an autonomous 'Refactor Agent' that auto-accepts file edits for routine cleanup tasks