This handles the hard part of visual testing: making sure screenshots actually capture rendered content before you analyze them. The core insight is smart: traditional pixel comparison tells you *what* changed, but LLM vision analysis tells you *why* it matters. It's built around waiting for React hydration properly (not just network idle), running Playwright headless to avoid window spam, and then feeding screenshots to Claude with specific questions about theme compliance, broken layouts, or accessibility issues. Most useful when you're testing UI states that are annoying to verify manually, like dark mode switches or responsive breakpoints. The code patterns for content verification before screenshot capture are genuinely helpful.
npx skills add https://github.com/erichowens/some_claude_skills --skill playwright-screenshot-inspector