Can I audit just one specific agent?

Yes, you can provide an agent name as an argument (e.g., /audit-agents my-agent) to focus the audit on a specific subagent rather than the entire ecosystem.

Where does the skill look for agents to audit?

It scans multiple sources including local marketplace repositories, the project-specific '.claude/agents/' directory, and globally installed plugins in your user profile.

What do the audit scores mean?

Agents receive a numerical score: 85+ is a PASS (ready for use), 70-84 is PASS WITH WARNINGS (functional but needs improvement), and below 70 is a FAIL (requires immediate fixes).

How does it handle false positives in the audit?

Unless skipped, the skill runs a validation phase where a specialized 'audit-finding-validator' agent reviews all initial findings to filter out false positives before presenting the final report.

What specifically does this skill audit in a subagent?

It checks YAML frontmatter (name, description, tools, model), naming conventions (lowercase kebab-case), description quality (third-person guidance), tool access configuration, and whether the selected model (Haiku/Sonnet/Opus) matches the task complexity.

Agent Auditor

Name: Agent Auditor
Author: melodic-software

bymelodic-software

•

보안 및 테스팅

Audits Claude Code subagents for quality, compliance, and maintainability to ensure optimal performance and documentation standards.

The Agent Auditor skill provides a comprehensive quality assurance framework for Claude Code subagents. It automatically evaluates agent definitions against official documentation and best practices, analyzing critical metadata including YAML frontmatter, naming conventions, description quality, and model selection appropriateness. By utilizing specialized subagents to validate findings and eliminate false positives, it ensures that your agent ecosystem remains robust, well-documented, and efficient across local project files and global plugin installations.

주요 기능

01Smart discovery of agents across local, project-specific, and global plugin directories

02Detailed scoring system (PASS/FAIL) based on documentation and maintainability standards

03Sophisticated finding validation phase to filter out false positives and ensure accuracy

04Automated compliance checking for subagent YAML frontmatter and naming conventions

0538 GitHub stars

06Persistent audit logging and recovery capabilities for interrupted sessions

사용 사례

01Verifying subagent quality and tool configurations before a production release

02Standardizing descriptions and naming conventions across a team's shared agent library

03Conducting periodic maintenance on large agent ecosystems to identify stale or non-compliant agents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add melodic-software/claude-code-plugins audit-agents

For use in Claude.ai and ChatGPT

Download Skill