SuperAPI AI Directory

ux-collab-setup

Testing & quality

View repository

Updated 10-15 23:48

About this skill

UX Collab Setup Skill

Diagnose and fix missing ux-collab dependencies in one guided session.

Step 1 — Run the check

Find the package root and run the check script:

# From the cloned repo / package directory:
npm run check      # or: pnpm run check
# or:
bash scripts/check.sh

Read the output carefully. Each line is either ✔ (pass), ⚠ (warning/optional), or ✘ (required — must fix).

Step 2 — Auto-fix what can be automated

Run the setup script — it handles ImageMagick installation, agent-browser installation, and creates .ux-collab.md if missing:

# From the cloned repo / package directory:
npm run setup      # or: pnpm run setup
# or:
bash scripts/setup.sh

This will:

  • Check if agent-browser is installed; if not, attempt to install via brew (macOS) or prompt for npm
  • Install ImageMagick via brew (macOS) or apt (Ubuntu/Debian) if missing
  • Make optimize-screenshot.sh executable
  • Copy optimize-screenshot.sh to ~/.agents/skills/ux-collab/ if the skill is installed there
  • Create a starter .ux-collab.md in the current directory if none exists
  • Create a sample agent-browser.json config if desired
  • Re-run check.sh at the end to confirm

Step 3 — Install agent-browser (PRIMARY browser tool)

agent-browser is the recommended browser tool for ux-collab — it's faster, token-efficient, and requires no MCP setup.

macOS (recommended via Homebrew — fastest)

brew install agent-browser
agent-browser install   # Downloads Chrome for Testing (first time only)

Any Platform (via npm)

npm install -g agent-browser
agent-browser install   # Downloads Chrome for Testing (first time only)

Verify Installation

agent-browser --version
# Should output: agent-browser x.x.x

agent-browser open https://example.com
agent-browser snapshot -i
agent-browser close

Optional: Project Configuration

Create agent-browser.json in your project root for persistent defaults:

{
  "headed": false,
  "screenshotDir": "./screenshots",
  "device": "iPhone 12"
}

See https://agent-browser.dev/configuration for all options.

Step 4 — Fix Playwright MCP (Alternative / Optional)

Playwright MCP is optional — agent-browser covers all core ux-collab needs. Only install if you need:

  • Full accessibility tree with semantic roles
  • Complex multi-page interactions with state
  • Specific viewport/device emulation
  • MCP ecosystem already configured and working

Claude Code (plugin system)

claude plugin install playwright@claude-plugins-official

Then restart Claude Code. Re-run npm run check (or bash scripts/check.sh) to confirm.

If MCP servers start but hang or show pnpm path errors

This is a known issue on WSL and macOS when pnpm is managed via fnm or a non-standard path. Use the appropriate fixer:

npx @kylebrodeur/mcp-wsl-setup   # WSL (fixes fnm/pnpm paths in VS Code mcp.json)
npx @kylebrodeur/mcp-mac-setup   # macOS (fixes pnpm paths)

Re-run npm run check (or bash scripts/check.sh) after.

Any harness (manual MCP server via .mcp.json)

Create or edit .mcp.json in the project root:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Restart the agent session. Re-run npm run check (or bash scripts/check.sh) to confirm.

Step 5 — Fix Lucid MCP (optional — skip if not needed)

Lucid MCP enables diagram creation in Step 3 of the ux-collab loop. Without it, the skill falls back to Markdown wireframes automatically — this is not a blocker.

To enable Lucid wireframes:

# Claude Code:
claude plugin install lucid@claude-plugins-official

# Or add to .mcp.json:
# "lucid": { "command": "npx", "args": ["@lucid/mcp@latest"] }

Step 6 — Create project config (if warned)

If check warned that .ux-collab.md is missing, create one by copying the template from the repo or create manually:

# From the cloned repo:
cp .ux-collab.md.example .ux-collab.md
# Then edit `.ux-collab.md` to set:
# - `defaultUrl` — your dev server URL
# - `lucidShareEmail` — your Lucid account email
# - Brand tokens, target files, and surfaces

Step 7 — Final verification

npm run check      # or: pnpm run check
# or:
bash scripts/check.sh

Expected output when fully ready:

✔ Passed:   6+
⚠ Warnings: 0–2  (Lucid and Playwright MCP are optional)
✘ Failed:   0
→ All required dependencies present. Ready to run ux-collab.
  Primary: agent-browser (installed)
  Alternative: Playwright MCP (optional)

If any ✘ remain, re-read the "Issues to fix" section printed by the script and follow the instructions there.

Quick reference — what each dep does

Dep Required? Why it's needed
agent-browser ✅ Yes Primary browser tool. Takes screenshots, navigates, interacts. Rust CLI, no MCP needed.
image magick convert + identify ✅ Yes Resizes and compresses screenshots before attaching to chat
Playwright MCP (mcp_playwright_*) ⚡ Optional Alternative browser tool. Use when richer accessibility features needed.
Lucid MCP (mcp_lucid_*) ⚡ Optional Creates and exports wireframe diagrams — Markdown fallback available
.ux-collab.md ⚡ Recommended Per-project config: dev URL, brand tokens, file targets, open decisions
agent-browser.json ⚡ Optional Project-level agent-browser defaults

Tool Selection Guide

Need Use
Quick screenshot + review agent-browser — one-liner, no MCP
Token efficiency (constrained context) agent-browser — ~200-400 tokens vs 3000-5000
CI/CD or terminal-only agent-browser — needs no MCP infrastructure
Headless local dev agent-browser — purpose-built for this
Full accessibility tree semantics Playwright MCP — richer semantic data
Complex multi-page interactions Playwright MCP — stateful sessions, richer API