This is a decision framework for designing loading states in AI interfaces. It breaks down waiting patterns into four categories (reasoning display, progress steps, streaming states, status indicators) and gives you specific guidance for each. The "elevator mirror effect" principle is solid: users perceive wait time as shorter when they have something to watch. The comparisons between Claude, ChatGPT, DeepSeek, and Gemini are particularly useful for understanding tradeoffs between transparency and cognitive load. Use this when you're building any AI feature where users wait more than a second or two and you need to decide whether to show thinking steps, hide them, or make them expandable.
npx skills add https://github.com/shipshitdev/library --skill ai-loading-ux