Skill authoring and evaluation

Workflow

Anthropic’s skill-creator workflow iterates on skill instructions, trigger descriptions and evaluations. The bundled evaluation workflow uses Claude tooling; install the scripts and reference material together.

Run this command in your project after reviewing the instructions and access terms. Install complete upstream directories to keep referenced rules and scripts.

npx skills add https://github.com/anthropics/skills/tree/dbd4588f9e1033efb41dad4bef2f7947c8993d44/skills/skill-creator --skill skill-creator --agent claude-code

Project file: .claude/skills/skill-creator/SKILL.md

This brief helps you choose the skill. Read the complete upstream SKILL.md and companion files for its actual instructions.

# Skill authoring and evaluation (anthropic-skill-creator)
Anthropic’s skill-creator workflow iterates on skill instructions, trigger descriptions and evaluations. The bundled evaluation workflow uses Claude tooling; install the scripts and reference material together.

## Use when
- Create a reusable agent skill or evaluate an existing skill.

## Required inputs
- Task definition and representative prompts
- Existing skill, expected outputs and evaluation criteria

## Prerequisites
- Claude Code plus Python and the bundled evaluation scripts for the documented evaluation workflow.
- Model calls used by evaluations may incur usage charges.

## Expected outputs
- Versioned SKILL.md package
- Evaluation results and revised trigger description

Source: https://github.com/anthropics/skills/blob/dbd4588f9e1033efb41dad4bef2f7947c8993d44/skills/skill-creator/SKILL.md
Full SKILL.md: https://agenttaste.dev/skills/anthropic-skill-creator/SKILL.md