How does Claude know when to use a custom agent?

Claude uses the 'description' field in the agent's frontmatter to auto-delegate tasks. Effective descriptions should use third-person triggers and identify specific domain keywords.

Which model should I use for my Claude agent?

The choice depends on complexity: Haiku is ideal for high-volume simple tasks, Sonnet for balanced general workflows, and Opus for critical reasoning and deep code analysis.

What are the common mistakes when creating agents?

Common errors include vague descriptions that fail to trigger delegation, providing too many unnecessary tools, and using first-person language in the agent definition.

What is a Claude Code Agent?

An agent is a specialized Claude instance spawned to handle specific responsibilities—like security audits or research—using a focused system prompt and restricted tools.

Why should I restrict tools for an agent?

Limiting tools to only what is necessary (e.g., Read-only for a reviewer) creates more focused behavior, improves token efficiency, and reduces the risk of unintended code modifications.

Claude Agent Creator

Name: Claude Agent Creator
Author: ed3dai

byed3dai

•

Developer Tools

Provides standardized patterns and best practices for building specialized, autonomous subagents within the Claude Code ecosystem.

This skill serves as a comprehensive framework for developers looking to extend Claude Code's capabilities through custom agents. It guides users through the technical requirements of agent creation, including writing effective auto-delegation descriptions that trigger based on specific symptoms, optimizing tool selection to ensure focused behavior, and implementing structured system prompts. By leveraging these patterns, developers can build domain-specific agents—such as security auditors, researchers, or implementors—that handle complex workflows with context isolation and high reliability.

Key Features

01Role-specific tool restriction strategies to enhance focus and security

025 GitHub stars

03Standardized testing protocols for verifying agent delegation and output

04Auto-delegation logic for seamless agent triggering and task matching

05Model selection optimization (Haiku, Sonnet, Opus) based on task complexity

06Standardized prompt templates for roles, responsibilities, and workflows

Use Cases

01Creating a TDD-focused implementation agent that follows strict testing workflows

02Developing a research subagent that synthesizes documentation from multiple web sources

03Building a specialized code reviewer agent with restricted read-only permissions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ed3dai/ed3d-plugins-testing creating-an-agent

For use in Claude.ai and ChatGPT

Download Skill