What Breaks — Design agent for Claude Code
Opus 5 validator that pressure-tests a handed artifact — diff, PR, document, design, claim, or question — and returns an evidence-backed assessment of what fails.
How to install What Breaks
Installs to ~/.claude/agents/ryanthedev-what-skills-what-breaks.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/ryanthedev/what-skills/HEAD/agents/what-breaks.md -o ~/.claude/agents/ryanthedev-what-skills-what-breaks.md Restart Claude Code, or start a new session, for it to be picked up.
What What Breaks does
name: what-breaks description: "Opus 5 validator that pressure-tests a handed artifact — diff, PR, document, design, claim, or question — and returns an evidence-backed assessment of what fails. Never edits." model: opus tools: Read, Grep, Glob, Skill
Validator brief
You are an independent validator: your job is to find what makes the artifact in front of you fail, and to say so plainly — including when the honest answer is that nothing does. Your assessment is relayed largely verbat
Alternatives in Design
- UI Visual Validator — You are an experienced UI visual validation expert specializing in comprehensive visual testing and 31.9k ★
- Design Validator 597 ★
- Chief — Head of the PO Council and eight-seat design bench 79 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Design Reviewer Facts
The evidence reviewer of the stage-2 design review round — every claim about an external tool or internal serv
Second Opinion Codex
Cross-vendor second opinion from the OpenAI Codex CLI. Use ONLY when the user explicitly asks for a codex/Open
Second Opinion Gemini
Cross-vendor second opinion from the Google Gemini CLI. Use ONLY when the user explicitly asks for a gemini/Go
Game Critic
Adversarial validator for game-design proposals. Finds counter-examples, known failure cases, and post-mortem
Dnp Performance Analyst
⚡ .NET performance analysis — async hotspots, N+1 queries, missing caching opportunities, allocation pressure,
Codex Deep Reasoner
CLI-backed read-only deep-reasoning worker — the codex variant of deep-reasoner. Use for root-cause analysis,
Related Skills
Aegis Return Format
Sub-agent return format — every non-trivial claim must be tagged [VERIFIED: ] (backed by executed command) or
Execute Right
A Claude Code skill for the moment a plan is carried out: a handed value is a hypothesis, a done-claim names i
Sitemend
SEO regression tests for your repo. Fails the PR when a change breaks crawling or indexing (preview URLs in ca