vmarafetti/crit ? reverse-engineered prompt

Build me a Claude Code skill called crit that audits a URL and gives a plain language design review from six angles, accessibility, motion, UX heuristics with dark pattern detection, UI baseline, AI interaction, and agent readability.

I want two modes, a full audit that writes a markdown report and a quick scan that returns only the top 5 critical issues in chat. The report should rank problems by severity and effort, tag each finding for PM, designer, engineer, or founder, and explain what is happening, why it matters, and what to do next in simple language. Include a short executive summary, a prioritized fix list, and deeper notes for each lens. Also include clear limitations so it says what was not checked.

Make it feel like a real audit tool for founders and product teams, not a compliance robot. If you need current docs or standards details, look them up online. I’d like the skill to be easy to install in Claude Code and to include a framework reference command that explains the method without running an audit.