What is a Claude Code sub-agent?

A sub-agent is a specialized subprocess within Claude Code that runs in an independent context to perform specific tasks using its own dedicated tools, models, and instructions.

Do sub-agents inherit skills automatically?

No, sub-agents do not automatically inherit skills from the parent. You must explicitly list the skills the agent needs to access in the 'skills' field of its frontmatter.

How do I define tool access for a sub-agent?

Tool access is defined in the frontmatter's 'tools' field. If omitted, the agent inherits all tools from the parent. If specific tools like Read, Glob, or Bash are listed, the agent is restricted to only those tools.

Which model should I specify for my sub-agent?

The 'inherit' value is recommended to match the user's current model. Use 'haiku' for fast, routine tasks and 'opus' for tasks requiring the highest level of reasoning and precision.

What are the mandatory fields for an agent file?

Every agent Markdown file must include a YAML frontmatter block containing at least the 'name' (identifier) and 'description' (explaining when the agent should be invoked).

Claude Code Agent Architect

Name: Claude Code Agent Architect
Author: biwakonbu

bybiwakonbu

開発者ツール

Provides technical specifications and implementation patterns for building and maintaining modular Claude Code sub-agents.

概要

The Agent Spec skill serves as a comprehensive reference and guide for developers building sub-agents within the Claude Code ecosystem. It provides deep knowledge of the agent lifecycle, including correct frontmatter configuration for required fields like name and description, as well as optional parameters for tools, model selection, and skill integration. By offering standardized templates and validation rules, this skill ensures that developers can implement robust, high-performance agents for specialized tasks such as code reviews, security auditing, and automated testing while adhering to the latest Anthropic specifications.

主な機能

Detailed guidance on tool inheritance and permission management
Best practices for kebab-case file organization and directory structure
0 GitHub stars
Optimized model selection advice (inherit, haiku, opus) for task efficiency
Instructions for cross-referencing and manually inheriting custom skills
Standardized frontmatter templates for rapid agent initialization

ユースケース

Implementing security-focused agents that leverage specific auditing skill sets
Building task-specific execution agents for running tests and parsing CLI output
Creating specialized sub-agents for automated code reviews and quality assurance

概要

主な機能

Detailed guidance on tool inheritance and permission management
Best practices for kebab-case file organization and directory structure
0 GitHub stars
Optimized model selection advice (inherit, haiku, opus) for task efficiency
Instructions for cross-referencing and manually inheriting custom skills
Standardized frontmatter templates for rapid agent initialization

ユースケース

Implementing security-focused agents that leverage specific auditing skill sets
Building task-specific execution agents for running tests and parsing CLI output
Creating specialized sub-agents for automated code reviews and quality assurance