smokeeaasd/flawless-ui-ux

What it does

flawless-ui-ux is a skill for coding agents that steers interface work around product context, content, and a distinct visual identity instead of generic patterns. It covers four modes: creating a new interface, redesigning an existing one, refining visual character, and reviewing a page for readability, navigation, and accessibility. A companion CLI collects project context, inspects source files or a running page, and compares results after changes; the agent still makes the edits.

The skill installs into a range of agent hosts, including Codex, Claude Code, Cursor, GitHub Copilot CLI, OpenCode, Kiro, and others listed in the README. Node 22.12+ is the runtime requirement for the CLI.

Quick start

From a project directory, install the skill with npm, pnpm, or Bun:

npx flawless-ui-ux@latest install

In the terminal menu, choose project or global scope, select one or more agents, and confirm the destinations. For scripted installation, pass flags directly:

npx flawless-ui-ux install --codex
npx flawless-ui-ux install --claude-code --global
npx flawless-ui-ux install --project ./app

--yes skips the menu and uses defaults: all supported agents in the current project. Then start a new agent session and refer to the skill by name, for example:

Use the flawless-ui-ux skill to build a financial dashboard for freelancers.

To inspect a running page, install Chromium once, capture a baseline, and verify after corrections:

npx playwright@1.64.1 install chromium
npx flawless-ui-ux inspect --project . --source src/App.tsx --url http://localhost:5173 --output ../app-before --format json
npx flawless-ui-ux verify ../app-before/run.json --output ../app-after --format json

The comparison separates persistent, new, and no-longer-observed findings.

Inspection scope

The CLI samples two viewports by default, 1440×900 and 390×844, and can take others with repeated --viewport flags. It checks selected axe contrast and accessible-name rules, computed font sizes and horizontal overflow, broken images, uncaught JavaScript errors, repeated copy, and arrow/icon cues in selected files. It does not establish full WCAG conformance, complete keyboard journeys, or visual quality, and it does not edit files, start servers, or download browsers itself.

Exit codes distinguish completed coverage without confirmed errors (0), completed coverage with confirmed errors (1), and invalid input or incomplete requested coverage (2). Use --selector to narrow inspection to a region and repeat --source for more files.

Installation notes

Reinstalling preserves existing files, local changes, and links, so review or remove an old installation first. The twelve supported agents share five project copies of the skill: hosts that discover .agents/skills use the shared copy, while others use native directories. Grok Build requires folder trust before project skills appear, and DeepSeek Harness should be installed from the nearest Git root. The exact paths and compatibility evidence are documented in the installation contract and testing guide. The Bash installer defaults to global scope and requires --project for local installation; on Windows it needs Git Bash or WSL.

HotRepo's take

The repository recorded 101 stars on November 8, 2026, rising to 112 by that evening, with one new-listing snapshot showing 11 stars and 1 fork gained in a five-hour window. It has no recorded best rank on the all-language daily board, so the available data is too thin to treat it as a trending project.

The more useful signal is the discussion around the announcement. A commenter compared this skill against two similar "de-slop" skills with example builds for the same landing-page prompt and judged that the one-shot results get you to a starting point rather than a finished design examples and comparison. Others asked for a demo before trying it and questioned how a design skill can be evaluated at all discussion. The author responded that examples were being requested reply, and the repository already points to manual evaluations and a workflow guide as the place to see the method. That makes the evaluation material, not the star count, the thing to read before adopting it.