Free download
A UX audit skill for Claude
Point it at a screen or a whole user journey. It asks what the user is trying to do, walks every screen against fifteen-plus usability frameworks, and gives you severity-rated findings with the heuristic cited, pinned to the exact region of the screenshot.
Email only, no name, no company. I'll send you the file and a few notes on using AI for deeper design work. Unsubscribe anytime.
What you get back
Three files, one source of truth
The report and the interactive preview are both generated from the structured findings, so they cannot drift apart.
audit.html
Open this one. Your screenshots with numbered pins on them. Click a pin to jump to the finding, tick the findings you want to act on, copy the shortlist.
audit-report.md
The written report. Findings grouped by screen, then journey-level issues.
findings.json
The structured output everything else is generated from. Every finding with its heuristic, severity, and screen region.
The rubric
What it audits against
Most AI design feedback is generic because the request is generic. This asks for evaluation against named, established principles, so findings arrive with a citation attached and are much harder to wave away in a review.
Classic usability
Nielsen's 10, Shneiderman's 8 golden rules, Gerhardt-Powals' cognitive engineering principles, Bastien & Scapin's ergonomic criteria.
Behavioural laws
Fogg Behaviour Model, Cialdini's principles of influence, peak-end rule, Hick's Law, Fitts's Law, Miller's Law, Jakob's Law.
Interaction and layout
Gestalt principles, Norman's design principles, Tognazzini's first principles, Krug's laws, progressive disclosure, F-pattern and Z-pattern scanning.
Accessibility
POUR and the WCAG criteria you can check visually. Anything needing a keyboard, screen reader, or motion check is flagged for manual follow-up rather than guessed at.
Dark patterns
Brignull's taxonomy, so a manipulative pattern gets named precisely.
Three custom check sets
FTU for first-time users, TRUST for high-stakes moments, COG for cognitive load. Not formal frameworks: named categories for the issues that kept recurring.
Honest limits
What it isn't
Not user research
It tells you where problems probably are. Users tell you what actually confuses them. The audit is where you start, not where you finish.
Not a full accessibility audit
It catches what is visible in a static image and is explicit about what still needs a keyboard, screen reader, and motion pass.
Not a prioritisation engine
It produces a list. Deciding which findings to fix, and in what order, given your users and what your team can ship, is still your job.
Get the skill
To install it, open the downloaded file with Claude. Then type /ux-audit and point it at a folder of screenshots. On claude.ai, upload it under Settings, then Capabilities, then Skills.
Prefer to read first? The full method, including every FTU, TRUST and COG check, is written up in how to run a UX audit with AI. Nothing in the skill is hidden from that post.