How does this skill reduce context window costs?

It identifies 'token density' violations and suggests moving deterministic code blocks into scripts, ensuring only essential instructions occupy the active context.

Can this skill help with routing problems?

Yes, it specifically audits Dimension 1 (Frontmatter), checking for trigger phrase authenticity and description framing to ensure the model routes tasks to the skill correctly.

When should I use repair-skill instead of improve-skill?

Use repair-skill for structural fixes, auditing quality, and fixing errors in existing logic. Use improve-skill when you want to add new features or enhance the skill's overall effectiveness.

What is the 'gold standard' used for the audit?

The audit uses reference files for skill anatomy, frontmatter options, and calibration patterns to ensure every skill follows the highly opinionated standards of the Claudest ecosystem.

Skill Repair & Audit

Name: Skill Repair & Audit
Author: gupsammy

bygupsammy

•

开发者工具

Audits and repairs Claude Code skills by diagnosing structural violations and identifying gaps against established development standards.

The repair-skill is a specialized diagnostic tool designed to maintain the quality and architectural integrity of Claude Code skills. It performs a rigorous multi-phase audit of a skill's directory or SKILL.md file, evaluating it across seven critical dimensions including frontmatter quality, execution modifiers, and context efficiency. By comparing your skill against a gold standard for anatomy and instruction logic, it generates a structured improvement plan that identifies violations, gaps, and polish opportunities to ensure your skills are reliable, cost-effective, and correctly routed.

主要功能

01Multi-dimensional audit of frontmatter, instructions, and directory structure

02Automated detection of frontmatter violations and tool-call inefficiencies

03Validation of trigger phrase authenticity for improved routing accuracy

04Identification of deterministic script candidates to reduce context token waste

0552 GitHub stars

06Analysis of agentic vs. deterministic splits to ensure workflow reliability

使用场景

01Reviewing third-party skills for structural correctness and context efficiency

02Optimizing skill performance by moving redundant logic into scripts

03Diagnosing why a specific skill is failing to trigger or routing incorrectly

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gupsammy/claudest repair-skill

For use in Claude.ai and ChatGPT

Download Skill