skill-creator

Create, evaluate, and package Claude Skills with evals and benchmarks.

14|1|Updated Apr 6, 2026
One-click install
npx skills add https://github.com/stay-leave/mini_agent --skill skill-creator-stay-leave
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: skill-creator
Source: https://github.com/stay-leave/mini_agent/tree/main/skills/skill-creator
Command: npx skills add https://github.com/stay-leave/mini_agent --skill skill-creator-stay-leave

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

Create and refine Claude Skills end-to-end by guiding drafting, evaluation, and iteration, transforming a raw idea into a polished, publish-ready Skill with evals, benchmarks, and packaging.

Core Features & Use Cases

  • Guided SKILL.md authoring including frontmatter and rich instructions
  • Built-in end-to-end evaluation loop (run_eval, improve_description, run_loop) with train/test split
  • Benchmark aggregation and optional viewer-based review for objective scoring
  • Support for optional resources in scripts/, references/, and assets/ to power deterministic tasks and rich outputs
  • Easy packaging of skills for distribution and deployment

Quick Start

Run the end-to-end skill-creation loop starting from a draft SKILL.md to iteratively test, evaluate, and optimize the skill using evals, viewer, and benchmarking tools.

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 evaluate Claude Skills end-to-end?

To create and evaluate Claude Skills, you can run an automated loop that drafts SKILL.md files, executes train/test split evaluations, and iteratively optimizes instructions using benchmarks and viewer-based reviews.

What is the best way to automate skill creation and evaluation?

Automating skill creation involves using a structured workflow to draft instructions, run built-in evaluation loops, aggregate benchmark scores, and package the final output for distribution.

Can I use viewer-based reviews to benchmark Claude Skills?

Yes, you can use optional viewer-based reviews to objectively score and benchmark Claude Skills, facilitating structured milestone tracking and iterative improvement during the optimization process.

Do I need scripts and references to build deterministic tasks for Claude Skills?

You can optionally add resources in scripts, references, and assets directories to power deterministic tasks and generate rich outputs when creating and testing your Claude Skills.

How does the iterative evaluation loop work for skill creation?

The iterative evaluation loop works by running evaluations, improving the skill description based on results, and repeating the cycle using a train/test split to ensure the skill is polished and publish-ready.

What are the limitations of automated skill creation and benchmarking?

Automated skill creation relies on structured milestones and train/test splits, meaning it requires well-defined evaluation criteria to function effectively and may not suit tasks lacking deterministic outcomes.