ux-critique

Installation
SKILL.md

UX Critique

Evaluate how well a user can understand and complete their goal. Output a single report that ranks experience issues by impact on task success, classifies each finding, documents the actual walkthrough, and calls out what works.

This is the experience counterpart to ui-design-critique (visual design). Where that skill judges how it looks, this judges how it works: task success, friction, feedback, recovery, cognitive load, navigation, forms, and manipulative patterns. The two are complementary — run both for a full picture.

The fundamental difference: aesthetics are brand-relative, so ui-design-critique needs a stated intent (a DESIGN.md) to judge against. UX is not brand-relative. It is judged against universal usability truths that hold regardless of brand, aesthetic, or taste — a keyboard trap, a dead-end flow, a silent failure, and a deceptive cancel path are wrong everywhere. So this skill needs no design spec, DESIGN.md, or equivalent. The principles in references/ux-heuristics.md and references/dark-patterns.md are the standard. The only thing you establish per-critique is the scenario — the user and their goal — not a source-of-truth document.

Scope: this skill evaluates and reports. It does not implement fixes. For building/polishing interaction and motion feel, use design-engineering; for self-evidence/affordance authoring, ux-interface-design.

Non-Negotiables

  1. Judge against universal usability truths and the user's goal, not aesthetics or a spec. The heuristics and dark-pattern checks are objective standards that apply to any product. Establish the task the flow is meant to support first; a finding matters to the degree it blocks or slows that goal. Never treat a provided design/journey doc as the standard — it describes intent, and intent can itself be wrong.
  2. Walk the real flow. Do not critique UX from a single static screenshot when the flow is runnable. Step through it. See references/task-walkthrough.md.
  3. Dark patterns are a first-class, P0-capable check. Manipulation, forced friction, and deception harm the user even when "usable." See references/dark-patterns.md.
  4. Every finding needs evidence, a priority, and a classification. Point to the step/screen/element. No unranked, unclassified, or evidence-free claims.
  5. Report strengths too. Name what works so it is preserved.

Inputs

Installs
7
GitHub Stars
63
First Seen
Jul 13, 2026
ux-critique — stolinski/s-stack