testuiux

Autonomous visual UI/UX defect detector and responsive layout verification engine powered by Playwright and Qwen2.5-VL. Captures multi-viewport screenshots, recognizes layout bugs, generates structured report.json, and triggers a lightweight triage model (SLM) to produce an executive briefing directly consumable by primary coding agents.

Features

• Zero Manual Setup: Fully autonomous model lifecycle. Automatically detects local Ollama instances, boots the daemon in the background if closed, and pulls models on-demand.
• Multi-Viewport Automated Capture: Uses Playwright to render mobile (375x812), tablet (768x1024), and desktop (1440x900) viewports.
• Vision-Language QA Inspection: Evaluates screenshots with Qwen2.5-VL to spot element collisions, text overlap, truncation, unconstrained horizontal overflows, broken assets, and unreadable color contrast.
• Visual Defect Highlighting: Uses Pillow to render color-coded bounding boxes and severity badges.
• Structured JSON Reporting: Automatically exports report.json with coordinate boundaries, severity ratings, and CSS remedies.
• Dual-Model Autonomous Triage: Feeds raw audit data into a compact language model (qwen2.5:1.5b) to synthesize an executive briefing (agent_briefing.txt) for primary AI coding assistants.
• Antigravity Skill Native: Registered as a persistent skill (testuiux) invoked automatically whenever frontend UI changes are implemented.

Requirements

• Python 3.10+
• Playwright Chromium (playwright install chromium)
• Ollama (optional, auto-detected if installed)

Installation

Install directly from PyPI:

pip install testuiux
playwright install chromium

Usage

1. Autonomous Run (Zero Configuration)

Point testuiux to any URL or local HTML file. The tool automatically detects runtime environment and produces JSON audit data with agent briefings:

testuiux http://localhost:3000

Or on local HTML files:

testuiux demo/buggy_site.html

2. Export All Formats (JSON + Side-by-Side HTML Report)

testuiux http://localhost:3000 --format all --output-dir ./audit_results

Generated Artifacts

audit_results/
├── report.json             # Structured machine-readable defects data
├── agent_briefing.txt      # Executive briefing for primary AI agent
├── report.html             # Standalone side-by-side comparison report
├── screenshot_mobile.png   # Clean capture
└── annotated_mobile.png    # Highlighted defects capture

Defect Categories

License

Licensed under the Apache License, Version 2.0. See LICENSE for details.