Key takeaways
- A skill for Claude Code and Codex hunts the telltale signs of AI-made UI: purple gradients, glows, emoji labels and shouty copy.
- It audits the screen and proposes a direction before it edits anything, so you can steer early.
- It works with common stacks such as Next.js, React, Vue, Svelte, Tailwind or plain HTML and CSS, even from a live link.
- If a screen is already good, it says so and suggests small refinements instead of redesigning.
- The catches: image generation may use a paid service, story videos need extra tools, and reuse needs credit.
What it is
unslop is a skill for Claude Code and Codex that turns vibecoded, AI-looking UI into a designed interface. It was created by Damir Sokolovsky, is released under CC BY 4.0, and the code is on GitHub at heydamir/unslop.
The problem it solves
Anyone who has asked an AI to build a dashboard or landing page knows the look. The README names the symptoms: purple gradients, glows, emoji labels, shouty copy and uneven spacing.
The skill finds those first, then sets a direction from your product and fixes the screen pass by pass, which the project compares to a designer doing a redline review.
What you get
- No surprise edits: it audits the screen and proposes a direction before it changes anything.
- A fixed order of work: spacing, colour, icons, copy, type, layout, details, visuals and motion.
- Broad reach: it works with Next.js, React, Vue, Svelte, Tailwind or plain HTML and CSS, and can start from a live link.
- Honest feedback: if a screen is already good, it says so and suggests small refinements.
- Something to post: it can make a short square video of the screen being cleaned up.
The catches
- Cost: generated visuals can go through an image service. The project says it shows the prompt and asks before every paid generation, and that dashboards and apps rarely need one.
- Extra tools: the story video needs Playwright and ffmpeg.
- Licence: reuse, including commercially, is allowed as long as you credit the author and link back.
- Promotion: the README also advertises the author's own paid design service.
- Evidence: the README shows the author's own example frames, not independent results.
What we did not test
We did not install the skill, run it on any screen or judge the quality of its redesigns. Taste is subjective, and the results will depend on your product and your agent. Every claim here comes from a literal quote in the project's README or the GitHub API, and the feature descriptions are the project's own.
Verdict
Editors' view: this is for developers who ship quickly with AI and are unhappy with how generic the screens look. The audit-first approach is sensible because it lets you steer before anything changes. Judge the results on a copy of your project, not your main branch.
Sources
Need help adopting this, or have a software project in mind?
Tell us what you are building. The initial consultation is free. Email [email protected] or follow @zolnaltd.
Start a conversation →