Verify frontend work in a real browser with Claude in Chrome

The problem

Claude ships UI changes it never saw rendered — layout bugs, console errors, and broken flows slip through because the agent cannot browse.

The hack

Install Claude in Chrome, start with `claude --chrome` (or enable by default via `/chrome`), and ask Claude to open localhost, exercise the flow, and fix what it observes.

Why it works

Anthropic’s power-user guide calls verification the #1 tip and specifically recommends the Chrome extension for frontend work — same standard you’d hold a human engineer to.

Setup

1. Install the Claude in Chrome extension from the Chrome Web Store. 2. Run `claude --chrome` (or `/chrome` → enable by default). 3. Approve the claude-in-chrome skill on first use. 4. Prompt: `Open localhost:3000, submit the form with invalid data, and fix any console/UI issues you see.` 5. Let Claude iterate in a real tab that shares your login state.

Prompt

Open localhost:3000, try the checkout flow, check the console for errors, and fix anything broken.

Prerequisites

Claude Pro/Max/Team/Enterprise with /login (not API-key-only or third-party providers).

Source

Related

• → Build a verification-loop skill that must run tests before done
• → Append /simplify after edits for a multi-agent quality pass