← ClaudeAtlas

cognitive-diversity-evaluatorlisted

Evaluates a UI, flow, or information architecture against a panel of ~20 cognitively diverse simulated users instead of a single "universal user." Each evaluator is randomly sampled on empirically-grounded cognitive and perceptual dimensions (working memory, visual search, processing speed, cognitive style), predicts where that specific profile hits friction, and the skill reports both consensus problems and where profiles DIVERGE — surfacing who a design silently optimizes for and who it strains. Use whenever the user wants a usability review, heuristic evaluation, cognitive walkthrough, accessibility-through-perception audit, UX critique, or asks whether users will struggle with a flow or who a design excludes — even if they don't use these exact terms. Also trigger on "evaluate this UX," "will users struggle with this," "review this interface/flow/IA," "accessibility of this design," or a screenshot / Figma export / HTML handed over for design critique.
nataliatalmina/skills · ★ 0 · Web & Frontend · score 70
Install: claude install-skill nataliatalmina/skills
# Cognitive Diversity Evaluator ## What this skill does and why it exists Classic heuristic evaluation and cognitive walkthroughs treat "the user" as a single universal entity. But individual differences in cognition and perception — how much people can hold in mind, how fast they process, how they search a display, how they prefer information represented — are large, stable, and directly consequential for whether a design works. Averaging them away hides real failures. This skill instantiates a panel of ~20 evaluators, each a distinct cognitive/ perceptual profile sampled fresh on every run. Each evaluator walks the artifact and predicts, *from its own parameters*, where it hits friction. The skill then reports two things: the problems that show up across many profiles (robust, high-confidence usability issues), and the points where profiles **disagree** — where the design serves one kind of mind and strains another. **This is a hypothesis generator, not a substitute for empirical user testing.** It front-loads likely issues and reveals whom a design favors, so real testing can be targeted. Its outputs are model-based predictions, not user data. Do not let it overclaim, and say so in the report. ## Inputs Accept any of: a screenshot, a Figma export or frames, a live URL or HTML, a written description of a flow, or an IA / sitemap. Model whatever you're given as a sequence of screens, steps, and decision points, noting at each what the design demands of the user (what m