W
writing-skills

writing-skills

obra
2026-04-06

Use when creating new skills, editing existing skills, or verifying skills work before deployment


Introduction

The writing-skills skill provides a meta-framework for creating and evaluating other AI skills. It solves the issue of developers writing ineffective, bloated, or overly specific AI prompts by treating prompt engineering as a disciplined, test-driven engineering task.

Core Concept

“Writing skills IS Test-Driven Development applied to process documentation.” If you haven’t observed an AI agent fail at a task without a specific instruction, you shouldn’t codify that instruction into a skill.

Installation and Usage Guide

https://github.com/obra/superpowers

Workflow Analysis (SOP)

Read original SKILL.md

  • TDD Mapping: The skill directly maps TDD concepts to prompt engineering (e.g., Test Case = Pressure scenario with subagent, Production Code = SKILL.md).
  • Watch it fail (Baseline): Before writing any documentation, the author must run a baseline scenario to document the exact rationalizations the AI uses to bypass best practices.
  • When to create: Instructs the user to only create skills for broadly applicable techniques that aren’t intuitively obvious, explicitly advising against writing skills for one-off solutions or project-specific conventions.
  • SKILL.md Structure: Mandates strict formatting for new skills, such as restricting the description to third-person and starting it with “Use when…” to ensure clear triggering conditions.

Skill Design Evaluation

  • Strengths of SOP Integration: Brilliantly adapts the principles of Test-Driven Development (TDD) to the process of writing documentation and prompt engineering. Defines clear criteria for when to create a skill versus when to rely on standard practices.
  • Potential Limitations: Treating documentation writing exactly like software testing (requiring failing baseline scenarios first) may be overly heavy for documenting simple, straightforward utility patterns.
  • What Makes a Good Skill: It forces the author to empirically prove the value of a skill by verifying that an AI agent actually fails without it, ensuring that only genuinely necessary and effective constraints are codified into skills.