Apply RED-GREEN-REFACTOR cycles to create and test AI skills.
npx skills add https://github.com/obra/superpowers --skill writing-skills-obra
Please help me install this Agent Skill. Skill: writing-skills Source: https://github.com/obra/superpowers/tree/main/skills/writing-skills Command: npx skills add https://github.com/obra/superpowers --skill writing-skills-obra
What problem does it solves? This Skill guides the creation and editing of skills (process documentation) using a Test-Driven Development (TDD) approach. It ensures skills are robust, discoverable, and resist rationalization by agents under pressure, making them truly effective.
Skill-Name-With-Hyphens (letters, numbers, hyphens only).name and description only (max 1024 chars total).High-intent search queries and answers about installing and using this skill.
Test-driven development for skills means writing failing test scenarios first (RED phase), then building minimal skill definitions to pass them (GREEN phase), then closing loopholes by identifying and countering rationalizations (REFACTOR phase). This approach ensures skills are bulletproof and resist agent workarounds under pressure.
Rationalization resistance comes from explicit counters to workarounds, rationalization tables that anticipate "spirit vs. letter" arguments, and red flags lists built during REFACTOR cycles. Testing with subagents under combined pressures reveals these loopholes before deployment.
Claude Search Optimization requires YAML frontmatter with name and description fields (max 1024 chars), a description starting with "Use when..." plus specific triggers and symptoms in third person, and rich keywords matching search queries. Token efficiency and discoverability improve when descriptions target concrete failure scenarios users face.
A skill requires: name field with hyphens only (letters, numbers, hyphens), YAML frontmatter with description, structured SKILL.md sections, and concise triggering descriptions. RED-GREEN-REFACTOR cycles and baseline testing against agent failures satisfy technical compliance requirements.
Yes. The RED-GREEN-REFACTOR cycle applies to any skill creation workflow. The approach is domain-agnostic; the metadata example shows TDD application, but the rationalization-bulletproofing method works for any skill where agents might bypass intent under pressure or competing objectives.
Skipping REFACTOR leaves rationalizations unaddressed. Agents will exploit gaps between the skill's stated intent and edge cases not explicitly countered. Testing without REFACTOR cycles produces skills that work in baseline scenarios but fail under pressure, competing priorities, or creative reinterpretation.
seanGSISGTest skills with subagents to verify resistance to rationalization under pressure.
nimeshgurungApply the TDD cycle to test and refine AI agent skills.
nice-and-preciseApply test-driven development cycles to verify process documentation with subagents.
Come0520Apply Test-Driven Development cycles to create and refine AI skills.