basher83/agent-auditor

Overview

This skill creates well-formed subagent definitions for Claude Code, producing either plugin-distributed agents or personal .claude agents with correct frontmatter and structure. It guides naming, capability selection, invocation criteria, and provides ready-to-drop markdown templates and examples. Use it to add specialized subagents to plugins or to your local agent collection.

How this skill works

It asks whether the agent is for a plugin or personal use, collects the agent purpose and capabilities, and generates a kebab-case name and YAML frontmatter matching the required format. For plugin agents it includes description and capabilities array; for user agents it includes name, description, optional tools and model, and a system prompt guideline. It also outputs sections: capabilities, when-to-use criteria, and context/examples demonstrating invocation.

When to use it

• When you ask to create, write, or generate an agent or subagent for Claude Code
• When adding a specialized agent to a Claude Code plugin and you need correct frontmatter and capabilities
• When you want a personal .claude agent with name, tools, and model hints
• When you need invocation criteria and concrete examples so Claude knows when to call the agent
• When you want a ready markdown template with title, sections, and examples

Best practices

• Decide plugin vs personal agent up front — plugin agents require a capabilities array, user agents require a name and can list tools
• Keep agents narrowly focused: 3–5 concrete capabilities describing specific actions or outputs
• Provide clear invocation criteria so Claude can decide when to invoke rather than handle tasks itself
• Use kebab-case for agent filenames and identifiers (e.g., security-reviewer)
• Include real context/examples showing typical user requests and expected agent responses

Example use cases

• Create a plugin agent 'security-reviewer' that inspects code diffs for vulnerabilities and recommends fixes
• Generate a personal agent 'performance-tester' in .claude for designing and analyzing load tests with specified tools
• Add an 'api-documenter' agent to a plugin that extracts OpenAPI fragments and drafts endpoint docs
• Produce frontmatter and full markdown for an agent that Claude can invoke when a user asks for specialized legal or compliance checks
• Provide invocation criteria examples so Claude calls the agent only for security, performance, or documentation requests, not general questions