Image: GitHub
Why it mattersAI coding agents reach for the same ornaments every time. One shared skill cuts that noise before the pull request, not after it.
Unslop UI passed 222 stars on GitHub and ranked fourth on Trendshift's daily list on 18 November. The project is an agent skill that removes AI tells from generated UI. Felipe Basurto, the author, calls it "opt-in surface rules" for agents working on UI copy, charts and dashboards.
The repo opens with one line that names the problem. "Agents decorate everything they build." Every section of the README is a pattern the skill will strip. Together it names 16.
What it strips
The list is specific. Eyebrow labels above a heading. Decorative numbering like 01//, 02//. Subtitle repetition that restates the heading. Explanatory captions on a button that already says what it does. Design notes written for the reader. Short status strings. Redundant labels on a form field. Slogans. Flavor badges. Animated status dots. Decorative emojis or icons. Chart titles that describe what the chart shows. Reading instructions. Duplicate values printed twice. Chart scaffolding. Legends in place of direct labels on a line.
How an agent uses it
The skill is installed per editor. Claude Code calls it as /unslop-ui. Cursor also uses /unslop-ui. Codex uses $unslop-ui. An agent that follows the SKILL.md manual can apply the same rules without the slash command. The skill holds the rules in plain text, so a team can read and change them.
The repo is MIT licensed. It has 225 stars at the time of writing, four contributors and 29 topics tagged on it. The topics name agent-skills, ai-slop, claude-code, claude-skills, codex, cursor, ui and ux-writing, so the audience it targets is clear: a developer using a coding agent to build a web interface.
The author
Felipe Basurto is a solo developer. His GitHub account dates to March 2020 and holds 29 public repositories. He has 18 followers. The traction here is on the project, not the brand.
The fix moves to the agent
A team that releases a dashboard built by an agent sees the same ornaments every week. The pattern list in this repo reads like a senior reviewer's running set of complaints. In one skill, callable from the editor, it moves the fix into the agent's own output.
Two limits are plain. The skill treats the ornaments as always wrong. A design that wants an eyebrow label, or a flavor badge, has to turn the rule off for that page. And the rule set is Basurto's own view of what AI UI gets wrong. A team with a different view has to fork and edit.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.