UX Review Skill
A multi-agent UX review system that browses live web applications with Playwright CLI and produces a prioritized, actionable report for solo developers and small teams.
Prerequisites
Playwright CLI must be installed. If not available, run:
npm install -g @playwright/cli@latest
playwright-cli install-browser
playwright-cli install --skillsVerify with playwright-cli --help.
Inputs
The user provides:
- Target URL (required) — The application's entry point (landing page, login, dashboard, etc.)
- Key user flows (optional) — Specific paths to test, e.g. "sign up → create project → invite team"
- Target audience (optional) — Who uses this app and what they expect
- Focus areas (optional) — Which review dimensions matter most (see Agent Roster below)
- Context file (optional) — A markdown file with app description, known issues, constraints
If the user doesn't provide optional inputs, use reasonable defaults and note assumptions.
Agent Roster
Each agent has a specialized lens. All agents use Playwright CLI to interact with the live app.
| # | Agent | Focus | Reference |
|---|---|---|---|
| 1 | First Impression | Landing page clarity, value prop, CTAs, trust signals, load perception | agents/first-impression.md |
| 2 | Information Architecture | Navigation, page hierarchy, naming, discoverability, mental models | agents/information-architecture.md |
| 3 | User Flow | Task completion paths, friction, dead ends, error recovery, onboarding | agents/user-flow.md |
| 4 | Visual Design | Spacing, typography, color, component consistency, responsive hints | agents/visual-design.md |
| 5 | Accessibility | WCAG compliance, keyboard nav, ARIA, contrast, form labels, error states | agents/accessibility.md |
| 6 | Copy & Microcopy | Button labels, error messages, empty states, onboarding text, tone | agents/copy-review.md |
| 7 | Synthesizer | Reads all agent reports → produces prioritized action plan | agents/synthesizer.md |
Workflow
Phase 1: Setup
- Confirm the target URL and any user-provided context
- Ensure Playwright CLI is installed (
playwright-cli --help) - Open a browser session:
playwright-cli open <url> - Take an initial screenshot to confirm the page loaded:
playwright-cli screenshot --filename=landing.png - Capture a snapshot for structural analysis:
playwright-cli snapshot --filename=landing-snapshot.txt - Create the workspace directory for outputs
Phase 2: Discovery (Main Agent)
Before spawning review agents, do a quick site crawl to understand scope:
- Capture the landing page snapshot
- Identify primary navigation links from the snapshot
- Visit 3-5 key pages (follow nav links) and snapshot each
- Build a simple sitemap of discovered pages
- Save discovery results to
workspace/discovery.md
This discovery output is shared with all review agents as context.
Phase 3: Parallel Review (Sub-agents)
Spawn agents 1-6 in parallel. Each agent receives:
- The target URL
- The discovery file (
workspace/discovery.md) - Their specific agent instructions (from
agents/*.md) - The Playwright CLI reference (
references/playwright-guide.md) - The heuristics reference (
references/heuristics.md) - An output directory for their findings
Each agent:
- Opens the app in its own Playwright CLI session (
-s=<agent-name>) - Navigates through relevant pages
- Takes snapshots and screenshots as evidence
- Writes findings to their output file
If sub-agents are not available, run each review sequentially in the main loop.
Phase 4: Synthesis
After all reviews complete, the Synthesizer agent:
- Reads all 6 agent reports
- Deduplicates overlapping findings
- Categorizes by severity (Critical / High / Medium / Low)
- Ranks by impact-vs-effort (essential for solo devs)
- Produces the final report
Phase 5: Output
Generate the final deliverable as a markdown report following the template in references/report-template.md.
Save to the workspace and present to the user.
Playwright CLI Quick Reference for Agents
Each agent should use a named session to avoid conflicts:
# Agent opens its own session
playwright-cli -s=first-impression open <url>
# Navigate
playwright-cli -s=first-impression goto <url>
# Capture page structure (element refs, text, layout)
playwright-cli -s=first-impression snapshot --filename=<name>.txt
# Take visual screenshot
playwright-cli -s=first-impression screenshot --filename=<name>.png
# Interact with elements (using refs from snapshot)
playwright-cli -s=first-impression click <ref>
playwright-cli -s=first-impression type <text>
playwright-cli -s=first-impression fill <ref> <text>
# Check navigation
playwright-cli -s=first-impression tab-list
# Close when done
playwright-cli -s=first-impression closeRead references/playwright-guide.md for the full command reference.
Output Structure
workspace/
├── discovery.md # Site structure from Phase 2
├── screenshots/ # Visual evidence
│ ├── landing.png
│ ├── nav-*.png
│ └── ...
├── agents/
│ ├── first-impression.md
│ ├── information-architecture.md
│ ├── user-flow.md
│ ├── visual-design.md
│ ├── accessibility.md
│ └── copy-review.md
└── ux-review-report.md # Final synthesized reportCoordinator Responsibilities
- Handle Playwright CLI installation if needed
- Run the discovery phase to build shared context
- Spawn review agents in parallel (or sequentially if no sub-agents)
- Ensure each agent uses its own named session (
-s=<name>) - Collect all agent reports
- Run the synthesizer to produce the final report
- Clean up browser sessions:
playwright-cli close-all - Present the final report to the user
Customization
The user can customize the review by:
- Skipping agents: "Skip accessibility, focus on user flows and copy"
- Adding competitor URLs: Include comparison analysis
- Specifying user stories: "Test the checkout flow specifically"
- Setting priorities: "I care most about conversion, less about accessibility"
- Providing context: "This is a B2B SaaS for project managers"
Adapt the agent roster and instructions accordingly.