Why does this skill enforce a 500-line limit?

To preserve the context window for actual tasks, skills should be concise. Longer documentation should use progressive disclosure through linked reference files.

What is Evaluation-Driven Development (EDD) for skills?

EDD involves creating specific success criteria and testing for a failure mode before writing the skill instructions, ensuring the instruction actually solves a real problem.

Does this skill work for all Anthropic models?

Yes, it includes a multi-model testing phase to ensure your custom skills perform correctly across Haiku, Sonnet, and Opus.

What are the pressure tests mentioned in the workflow?

Pressure tests evaluate if Claude will skip skill instructions when faced with simulated urgency, sunk cost fallacies, or direct user commands to ignore rules.

Skill Authoring & Evaluation

Name: Skill Authoring & Evaluation
Author: Veraticus

byVeraticus

0•

개발자 도구

Standardizes the creation and testing of Claude Code skills using a rigorous, evaluation-driven development workflow.

This skill implements a test-driven development (TDD) cycle specifically for authoring Claude Code skills, ensuring they are reliable, concise, and resistant to model 'rationalization.' It guides users through defining success criteria, establishing baseline failures (RED phase), creating minimal skill definitions (GREEN phase), and performing rigorous pressure tests against common loopholes like time constraints or authority pressure. By treating skills as executable documentation, it helps developers maintain high-quality prompt engineering standards while adhering to official Anthropic best practices for token management and progressive disclosure.

주요 기능

01Evaluation-driven development (EDD) workflow for robust skill creation

02Progressive disclosure patterns for managing complex documentation hierarchies

03Automated enforcement of Anthropic frontmatter and token budget limits

04Rigorous pressure testing against model shortcuts and rationalizations

05Multi-model validation suite for Haiku, Sonnet, and Opus models

060 GitHub stars

사용 사례

01Standardizing AI-assisted workflows across large-scale software repositories

02Developing custom team skills that require high consistency and reliability

03Refactoring existing skills to reduce context window bloat and improve performance

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add veraticus/gambit writing-skills

For use in Claude.ai and ChatGPT

Download Skill