erkamyaman/unslopovich

What it does

unslopovich is an agent skill plus a dependency-free JavaScript detector for spotting UI and UX patterns that make an app look or feel AI-generated. The detector implements 31 rules across SwiftUI, UIKit, Jetpack Compose, Android XML, React Native, Flutter, Ionic, React, Vue, Svelte, Astro, HTML, CSS and Tailwind classes. It catches items such as purple gradients, nested cards, decorative status dots, emoji used as icons, unvirtualized lists, disabled back gestures, hamburger menus in mobile apps, missing alt text, disabled zoom and removed focus rings.

The author's research claims to have read 1,201,466 App Store and Google Play reviews, of which 529,434 were rated 3 stars or lower. The analysis labels about a quarter of low-rating complaints as something a design and front-end pass can address. These are the author's findings, with method and limits documented separately in the study.

It is aimed at developers working with coding agents who want a structured review of an existing app rather than a one-off prompt. Flutter, SwiftUI, React Native, Compose, React and plain HTML are all in scope; the detector reads the nearest package.json to distinguish a phone app from a website.

Quick start

Requirements: Node.js 18 or later and a supported coding agent (Claude Code, Codex, Cursor, OpenCode, Gemini CLI or GitHub Copilot).

1. Install the skill and let the installer detect your agents:

npx unslopovich

2. In the agent, ask for a diagnosis:

/unslopovich review my app

Codex uses $unslopovich review my app; the other listed agents use the slash form per the install table.

3. To apply fixes, use the treat command with a target screen, for example:

/unslopovich treat checkout

You can also run the detector directly without an agent:

npx unslopovich detect src/
npx unslopovich detect --json ios/ android/

Exit code 2 means findings were reported, so it can gate a pull request. --list-rules prints every rule and its fix. The package is published on npm as unslopovich.

Usage notes

• The detector reads source code, not rendered screens. Problems that only appear at runtime, such as a popup in the way or a slow list, need the agent to inspect the running app.
• Reports mark performance issues as Not profiled; a clean review is not a performance or correctness guarantee.
• Rule suppressions: use // unslopovich-ignore <rule-id> for a single line (<!-- unslopovich-ignore <rule-id> --> in XML), or .unslopovich/config.json with ignoreRules and ignorePaths for project-wide rules.
• The detector guesses platform from the nearest package.json; override with --platform web or --platform mobile.
• The skill writes project context to .unslopovich/chart.md (top tasks and brand rules), and treats screens as healthy, light or heavy before making changes.
• Install is per user by default; --project installs into a project for the whole team. A manual copy of skills/unslopovich is also supported.
• License: MIT. No runtime dependencies in the detector.

HotRepo's take

unslopovich entered HotRepo's new JavaScript list on 2026-11-15 with 107 stars, up 3 in the recorded growth window, and 4 forks. The repository was created on 2026-11-09, so this is an early snapshot rather than an established trend. What stands out is the split between a deterministic detector and an agent-driven review: the CLI half works without any model, which gives the project a verifiable core beyond the skill prompt, and the suppression file and exit code make it usable in CI. The broader claims about how often users complain about design problems rest on the author's own review-labeling study; readers should check its method and sample limits before treating the percentages as general evidence.