skill-creator

Create and refine Claude skills by drafting SKILL.md and running iterative evals.

1|Updated Mar 12, 2026

Search Tags:#automation #documentation #evaluation #iteration #benchmark #skill

Ready to run

npx skills add https://github.com/mahou-labs/shiru --skill skill-creator-mahou-labs

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: skill-creator
Source: https://github.com/mahou-labs/shiru/tree/main/.agents/skills/skill-creator
Command: npx skills add https://github.com/mahou-labs/shiru --skill skill-creator-mahou-labs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill unit provides a structured framework for designing, testing, and iterating Claude skills end-to-end, including drafting SKILL.md, running evals, benchmarking, and describing improvements.

Core Features & Use Cases

End-to-end skill creation workflow: from interview to final SKILL.md and optional resources.
Built-in evaluation loop: run prompts, collect results, and automatically propose improvements.
Benchmark-ready: produce JSON benchmark data and human-readable reports to compare iterations.
Documentation & packaging tools: scripts to validate, package, and publish skills.

Quick Start

Use this workflow to create or improve a skill: draft the SKILL.md, run the evaluation loop on a sample eval set, review results, and iterate until you reach a stable, high-quality skill description.

Frequently Asked Questions about skill-creator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I create and iterate on Claude skills end-to-end?▼

To create Claude skills, you draft a SKILL.md file, run evaluation loops on sample sets, review benchmark results, and iterate until reaching a stable, production-ready skill description.

What is the best way to benchmark and evaluate skill triggering accuracy?▼

Benchmarking skill triggering accuracy involves running prompts through an evaluation loop, collecting JSON benchmark data, and reviewing human-readable reports to compare iterations and propose improvements.

Can I package and publish skills with bundled scripts and references?▼

Yes, skills can be packaged and published by bundling optional scripts, references, and assets for on-demand execution and verification to ensure reproducible outputs.

How does the skill creation workflow handle user interviews and drafting?▼

The skill creation workflow collects user interviews, drafts the initial SKILL.md file, and orchestrates iterative evaluations to produce a polished, production-ready skill description.

Do I need prior benchmark data to start refining an existing skill?▼

No prior benchmark data is required; the workflow generates JSON benchmark data and human-readable reports by running evaluation loops on a sample eval set to measure and compare iterations.