🔥 Claude Roaster
Brutal UI/UX critique with multi-perspective analysis - No mercy, maximum improvement.
A Claude Code plugin that roasts your UI/UX from multiple professional perspectives, with iterative improvement cycles and comprehensive user flow analysis.
What is This?
Ever wanted Gordon Ramsay to review your UI? This is it. The Roaster agent tears apart your interface with surgical precision, then helps you rebuild it better.
Features:
- 🎯 Multi-perspective analysis - 5 core specialists + 6 domain experts
- 🔄 Iterative improvement - Roast → Fix → Roast → Fix (1-10 configurable cycles)
- 📸 Platform-aware screenshots - Xcode MCP, Playwright MCP, or manual upload
- 🗺️ User flow analysis - Multi-screen journey analysis with friction detection
- 📊 Detailed reports - Markdown reports with linked screenshots
- 🎚️ Focus areas - Emphasize specific perspectives (accessibility, conversion, etc.)
- 🔌 Lifecycle hooks - Extensible hook system for custom integrations
- 🔥 Brutal honesty - No sugar-coating, but always actionable
Installation
# Via Claude Code plugin marketplace
/plugin marketplace add https://github.com/muspelheim/claude-roaster
/plugin install claude-roaster
Usage
Command Syntax
/roast [mode] [target] [options]
Modes:
screen - Single screen analysis (default)flow - Multi-screen user journeyaudit - Full application review
Options:
--iterations=<1-10> - Roast cycles (default: 3)--focus=<area> - Priority: a11y, conversion, usability, visual, implementation--output=<path> - Output directory (default: reports/roast/)--fix=<mode> - Fix handling: auto, report, ask (default: ask)
Examples
# Screen mode - single view analysis
/roast screen login
/roast screen settings --focus=a11y
/roast screen checkout --iterations=5
# Flow mode - user journey analysis
/roast flow checkout
/roast flow onboarding --fix=auto
# Audit mode - full app review
/roast audit
/roast audit --output=./audit-results
What Happens
- Capture - Screenshot taken via detected MCP
- Analyze - Agents run in parallel (see progress with
ctrl+o)
- Report - Results synthesized and saved to markdown
- Fix - Choose how to handle issues (asked after results shown)
- Iterate - Repeat for configured iterations
Agent progress is shown in Claude Code's native UI - press ctrl+o to expand the task tree and see each agent's status.
Smart defaults: Starts immediately. Fix preferences asked only after results.
How It Works
Roast Modes
| Mode | Command | What Gets Analyzed | Screenshots |
|---|
| Screen | /roast screen login | Single screen/view | 1 per iteration |
| Flow | /roast flow checkout | Multi-screen journey | 1 per step |
| Audit | /roast audit | Full app review | Multiple key screens |
Screen Mode: Analyzes one screen in depth. Best for focused improvements on specific views (login, settings, profile).
Flow Mode: Analyzes a user journey across multiple screens. Captures screenshots at each step, checks cross-screen consistency, identifies drop-off points.
Audit Mode: Comprehensive review of the entire application. Identifies critical screens automatically, analyzes key user paths.
MCP Detection
Claude Roaster automatically detects available screenshot tools in this order:
1. Check: Is Xcode MCP available?
└─ Yes → Use mcp__xcodebuildmcp__screenshot (iOS/macOS simulator)
└─ No → Continue...
2. Check: Is Playwright MCP available?
└─ Yes → Use mcp__playwright__browser_take_screenshot (web browser)
└─ No → Continue...
3. Fallback: Request manual screenshot upload from user
How detection works:
- Checks Claude Code's available MCP tools at session start
- No configuration needed - just have the MCP server running
- Falls back gracefully if no automation available
Requirements by platform:
| Platform | MCP Required | Setup |
|---|
| iOS/macOS | xcodebuildmcp | Simulator running or device connected |
| Web | Playwright MCP | Browser context active via browser_navigate |
| Any | None (manual) | User uploads screenshots when prompted |
Analysis Scope
What gets analyzed: