What is the MUST/SHOULD/NEVER format in Claude skills?

It is a specific syntax used to categorize rules: MUST for non-negotiable requirements, SHOULD for strong recommendations, and NEVER for absolute prohibitions, ensuring Claude follows instructions precisely.

How do I improve skill discovery for Claude?

Write descriptions in the third person focusing on 'when' to use the skill and specific symptoms or triggers, rather than summarizing the internal workflow.

When should I move content to a separate file in a skill?

If a reference section exceeds 100 lines or the main SKILL.md body approaches 500 lines, you should use progressive disclosure by moving details into separate markdown files.

What is 'Rationalization Prevention' in skill design?

It is a technique used in Discipline Skills to prevent Claude from making excuses to skip steps, often using an 'Iron Law' and a table comparing common excuses to reality.

Skill Authoring Guide

Name: Skill Authoring Guide
Author: jagreehal

byjagreehal

0•

Aprendizaje y Documentación

Standardizes the creation, editing, and optimization of Claude Code skills through structural patterns and discovery-focused best practices.

The Skill Authoring Guide provides a comprehensive framework for developing high-quality Claude Code skills that are easily discoverable and highly effective. It focuses on maximizing context efficiency by challenging every token and offers specific templates for different skill types, including reference guides, technical patterns, and discipline-enforcement rules. By utilizing the MUST/SHOULD/NEVER format and 'Iron Law' principles, this skill ensures that Claude remains compliant with complex workflows while preventing common AI rationalization errors during execution.

Características Principales

01Rationalization prevention through Iron Law and Red Flag tables

02Standardized MUST/SHOULD/NEVER rule formatting for clarity

03Guidelines for progressive disclosure and token efficiency

04Optimized discovery patterns for skill descriptions and triggers

050 GitHub stars

06Tailored frameworks for Reference, Technique, and Discipline skills

Casos de Uso

01Converting complex documentation into actionable Claude-ready formats

02Auditing and optimizing existing skills for better model adherence

03Creating new specialized Claude skills from scratch

Características Principales

01Rationalization prevention through Iron Law and Red Flag tables

02Standardized MUST/SHOULD/NEVER rule formatting for clarity

03Guidelines for progressive disclosure and token efficiency

04Optimized discovery patterns for skill descriptions and triggers

050 GitHub stars

06Tailored frameworks for Reference, Technique, and Discipline skills

Casos de Uso

01Converting complex documentation into actionable Claude-ready formats

02Auditing and optimizing existing skills for better model adherence

03Creating new specialized Claude skills from scratch