playwright-cli

name	playwright-cli
description	Authenticated browser automation for Synthex — drive real Chrome to log in, screenshot, and audit dashboard/integration surfaces. Use for live visual verification, "audit the dashboard", "check what's connected", or closing the verification gate on /dashboard/* routes. Two backends: the Playwright MCP server (interactive, in-session tools) and the committed CLI scripts (scripts/browser/*.mjs, reliable + headless). RUNS WHERE THE BROWSER + NETWORK + AUTH EXIST — i.e. a local machine, not a network-restricted sandbox.
metadata	{"author":"synthex","version":"1.0","type":"action-skill","triggers":["audit the dashboard","check what's connected","browser audit","visual verification","screenshot the dashboard","log in and check"]}
context	fork

Playwright CLI / Browser Automation

When to use

Live, authenticated checks of synthex.social surfaces: which integrations are connected, empty/error states, visual regressions — anything that needs a real logged-in browser. Pairs with browser-auth, browser-verify, site-smoke-test.

Hard reality (read first)

Browser automation runs in the environment it's launched from. A network-restricted Claude sandbox (no outbound npm, no auth session) cannot reach the authenticated app — run this on a machine with: network access, Playwright/Chromium installed, and the ability to sign in. Google SSO accounts (Continue with Google) cannot be driven headlessly; use a persistent profile + manual sign-in once, or a dedicated email/password test account.

Backend A — Playwright MCP (interactive, in claude sessions)

Wired in .claude/settings.json → mcpServers.playwright (npx @playwright/mcp@latest). Loads at Claude Code startup; restart the session to pick it up. Key tools:

• mcp__playwright__browser_navigate — open a URL
• mcp__playwright__browser_snapshot — accessibility/DOM snapshot
• mcp__playwright__browser_take_screenshot — capture
• mcp__playwright__browser_click / browser_type — interact (e.g. sign-in)

For SSO: run headed, complete Continue with Google by hand once; the MCP keeps the context for subsequent navigations in that session.

Backend B — committed CLI scripts (reliable, scriptable)

No MCP needed; uses the repo's Playwright.

# SSO (Google) accounts — capture a session once (real Chrome, persistent profile):
node scripts/browser/capture-session.mjs https://synthex.social
# then audit (reuses the profile, headless):
node scripts/browser/dashboard-audit.mjs https://synthex.social

# email/password accounts:
node --env-file=.env.local scripts/browser/dashboard-audit.mjs https://synthex.social
#   (set SYNTHEX_TEST_EMAIL / SYNTHEX_TEST_PASSWORD — a dedicated test account)

Output: screenshots in .artifacts/browser-audit/ + a JSON report (connected / not-connected / empty / error per surface). Exit: 0 ok · 2 no auth · 3 auth failed/expired · 4 launch error.

One-time setup: npm install · npx playwright install chrome.

Pass / fail

PASS = the audit reaches the dashboard and reports each surface's state with no unexpected error signals. FAIL = auth blocked (fix creds / use email-password account) or a surface shows an error/empty state that shouldn't be there → triage into a follow-up issue.

Reference

• scripts/browser/capture-session.mjs, scripts/browser/dashboard-audit.mjs
• .claude/skills/browser-auth/SKILL.md, browser-verify, site-smoke-test
• Playwright MCP: https://github.com/microsoft/playwright-mcp