What is a Claude Code Skill?

A specialized capability that provides domain-specific guidance, best practices, and implementation patterns to enhance Claude's behavior during specific tasks.

Can I use this for non-coding skills?

Yes. While it is built for developers, the logic of drafting, testing, and evaluating applies to any skill requiring structured instructions and verifiable outputs.

How does the Skill Creator help with undertriggering?

It helps refine the skill's YAML frontmatter and descriptions to ensure Claude recognizes exactly when the skill should be activated for the user.

What are the Building Blocks in this skill?

The skill uses four specialized agents: an Executor to run tests, a Grader to check expectations, a Comparator for blind A/B testing, and an Analyzer to suggest improvements.

Does it require subagents to function?

The Skill Creator can operate inline for sequential tasks, but it works most efficiently when it can spawn subagents to handle evaluation and benchmarking in parallel.

Skill Creator

Name: Skill Creator
Author: nablarch

bynablarch

0•

开发者工具

Streamlines the creation, optimization, and benchmarking of specialized Claude Code Skills through iterative evaluation and automated analysis.

Skill Creator is a comprehensive meta-skill designed to guide users through the entire lifecycle of developing custom Claude capabilities. It facilitates a structured workflow—from initial intent capture and drafting to rigorous automated evaluations, blind A/B testing, and post-hoc performance analysis. By leveraging composable building blocks like executors and graders, it helps developers refine skill triggers, improve output quality, and ensure consistent behavior across diverse use cases, ensuring that custom skills are both reliable and effective.

主要功能

01Post-hoc analysis to identify specific strengths and weaknesses

020 GitHub stars

03Iterative skill development loop (Draft, Test, Evaluate, Refine)

04Automated evaluation runs with transcript and metric capture

05Standardized benchmarking to measure skill performance at scale

06Unbiased blind comparison for A/B testing skill versions

使用场景

01Running regression tests on skills after updating instructions or reference files

02Developing a new specialized skill for a specific coding framework or internal workflow

03Optimizing an existing skill's trigger descriptions to prevent undertriggering

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nablarch/nabledge-dev skill-creator

For use in Claude.ai and ChatGPT

Download Skill