skills.rest
gavin100305
gavin100305COMMUNITY
Explore all skills by this author →
GitHub

skill-creator

Create and iteratively improve Claude skills with SKILL.md and benchmark evals.

1|Updated May 24, 2026
Search Tags:#skill creation#benchmarking#iteration#grading#evals#trigger optimization#LLM workflows
One-click install
npx skills add https://github.com/gavin100305/traffic --skill skill-creator-gavin100305
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: skill-creator
Source: https://github.com/gavin100305/traffic/tree/main/agents/.agents/skills/skill-creator
Command: npx skills add https://github.com/gavin100305/traffic --skill skill-creator-gavin100305

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pyyaml, xlsx, claude, python, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill helps you create new Claude skills from scratch and systematically improve existing skills so they trigger reliably and perform well on real test cases.

Core Features & Use Cases

  • Skill ideation to draft: Clarify intent, trigger conditions, expected outputs, and whether to add evals, then draft a complete SKILL.md.
  • Quantitative evaluation loop: Create test prompts, run them (with-skill vs baseline), draft assertions, grade results, and aggregate benchmarks to compare performance with variance.
  • Iterative improvement: Rewrite the skill based on qualitative feedback and quantitative benchmark patterns, then rerun and repeat until satisfied.
  • Description optimization: Generate trigger-eval queries and run an optimization loop to improve the SKILL.md frontmatter description so Claude triggers the skill at the right times.

Quick Start

Use this skill when you want to turn a workflow into a reusable Claude skill, then run a small test set, review the results, and iterate on the skill until it reliably triggers and produces the expected output.

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 from scratch?▼

To create Claude skills, draft a SKILL.md with frontmatter, generate eval prompts, run baseline comparisons, grade assertions, and use benchmark metrics to iteratively refine skill instructions and trigger behavior.

How does benchmarking and grading work for LLM workflows?▼

Benchmarking for LLM workflows works by executing with-skill versus baseline runs on test prompts, grading assertions against expected outputs, and aggregating metrics to compare performance with variance across multiple scenarios.

What's the best way to optimize skill trigger descriptions in Claude?▼

Trigger description optimization uses a held-out query set to evaluate and refine the SKILL.md frontmatter description, ensuring Claude triggers the skill reliably at the appropriate times without overfitting to test cases.

Do I need pyyaml and python to run skill creation and evaluation loops?▼

Yes, skill creation and evaluation loops require a Python environment with pyyaml for parsing SKILL.md frontmatter, along with xlsx and claude dependencies to execute benchmarking and aggregate results.

Can I use this skill to improve an existing SKILL.md rather than starting new?▼

Yes, you can improve an existing SKILL.md by running the quantitative evaluation loop against a baseline, analyzing benchmark patterns, and rewriting instructions or trigger descriptions to iteratively enhance output quality.

Why does my Claude skill fail to trigger reliably on real test cases?▼

Skills fail to trigger reliably when the frontmatter description is poorly optimized, requiring an iterative description optimization loop using held-out queries to refine trigger conditions and avoid overfitting.

Related Skills

Explore More Software Engineering Skills→
vuanhtu1993vuanhtu1993

skill-creator

Create, validate, and package Claude Skills into .skill files.

Community
Intermediate
rohunvorarohunvora
5

prompt-engineering

Design prompts and templates for LLM interactions and Agent commands.

Community
Advanced
MisLinkMisLink

skill-creator

Draft SKILL.md files, generate test prompts, and run with-skill versus baseline evaluations.

Community
Advanced
yahav123147yahav123147

skill-creator

Create and iterate SKILL.md structures with trigger evaluations and grading evidence.

Community
Advanced