Does this skill modify my agent files automatically?

It generates a structured improvement plan and diagnosis first, allowing you to review the recommended changes across seven dimensions before implementation.

How does the audit classify issues?

Findings are categorized into three types: Violations (contradicting rules), Gaps (missing improvements), and Improvements (polish), with severity levels ranging from minor to critical.

What is the difference between create-agent and repair-agent?

While create-agent generates new agents from scratch, repair-agent audits existing files to identify violations, missing features, and structural gaps against professional gold standards.

Can I use this skill to fix my Claude Skills?

No, this skill is specifically designed for agent files. If you need to audit or fix a skill, you should use the repair-skill command instead.

Agent Repair Specialist

Name: Agent Repair Specialist
Author: gupsammy

bygupsammy

•

セキュリティとテスト

Audits and optimizes existing Claude Code agents by diagnosing architectural violations and identifying quality gaps against gold-standard patterns.

The Agent Repair Specialist is a sophisticated diagnostic tool designed to elevate the performance and reliability of your Claude Code agents. Unlike generation tools, this skill performs a deep-tissue audit across seven critical dimensions—including description quality, frontmatter modifiers, and system prompt efficiency—to ensure your agents are structured for maximum routing accuracy and execution safety. It identifies critical violations, major architectural gaps, and minor polish improvements, providing a structured roadmap to transform basic scripts into production-grade autonomous agents.

主な機能

0152 GitHub stars

02Structured improvement plans categorizing findings by severity (Critical, Major, Minor).

03Automated detection of agentic vs. deterministic logic splits for better performance.

04Multi-dimensional audit covering description quality, frontmatter, and voice conventions.

05Context-aware identification of system prompt bloat and efficiency bottlenecks.

06Verification of routing triggers and XML example block completeness.

ユースケース

01Refining agent autonomy and safety by auditing tool access and frontmatter modifiers.

02Debugging a Claude Code agent that fails to trigger correctly or consistently during routing.

03Optimizing large system prompts by extracting domain data into specialized skill preloads.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gupsammy/claudest repair-agent

For use in Claude.ai and ChatGPT

Download Skill