Can I limit the tools an agent has access to?

Yes, you can define a specific allowlist in the agent's YAML frontmatter to restrict access to only necessary tools like Read, Write, or Bash for improved safety.

What is the difference between a Claude Agent and a Claude Skill?

Skills provide global guidelines that Claude follows automatically within a context, whereas Agents are specialized sub-instances launched explicitly for isolated, multi-step tasks.

Which model should I use for my custom agent?

Choose Haiku for simple, fast tasks; Sonnet for standard development workflows; and Opus for complex reasoning, architectural changes, or high-precision refactoring.

Where should I store my custom agent files?

User-level agents should be placed in the '.claude/agents/' directory, while plugin-specific agents reside in the 'agents/' folder of the plugin root.

How do I invoke a custom agent I've created?

Agents are launched using the Task tool by specifying the 'subagent_type' (matching your agent's filename) along with a detailed prompt and description.

Custom Claude Agents

Name: Custom Claude Agents
Author: vinnie357

byvinnie357

生産性とワークフロー

Enables the creation and configuration of specialized AI agents with restricted tool permissions and task-specific behaviors within Claude Code.

概要

The Claude Agents skill provides a robust framework for building autonomous sub-agents tailored to high-stakes or repetitive developer workflows. By leveraging Markdown-based configurations and YAML frontmatter, users can define specialized instances for tasks like security auditing, test generation, and code refactoring. Each agent can be optimized with specific tool allowlists, iteration limits, and model selections (Haiku, Sonnet, or Opus), allowing for isolated, multi-step execution that enhances the precision and safety of complex programming operations.

主な機能

Autonomous multi-step execution via integration with the Task tool
Markdown-based behavior definitions for easy versioning and documentation
Customizable tool allowlists to restrict agents to specific operations like Read or Grep
Flexible deployment via user-level or plugin-specific agent directories
Model-specific optimization to balance cost and reasoning capabilities per task
0 GitHub stars

ユースケース

Automating comprehensive security audits using a read-only analyzer agent
Performing safe, large-scale code migrations using high-iteration refactoring agents
Generating unit and integration test suites for new modules with a dedicated generator

概要

主な機能

Autonomous multi-step execution via integration with the Task tool
Markdown-based behavior definitions for easy versioning and documentation
Customizable tool allowlists to restrict agents to specific operations like Read or Grep
Flexible deployment via user-level or plugin-specific agent directories
Model-specific optimization to balance cost and reasoning capabilities per task
0 GitHub stars

ユースケース

Automating comprehensive security audits using a read-only analyzer agent
Performing safe, large-scale code migrations using high-iteration refactoring agents
Generating unit and integration test suites for new modules with a dedicated generator