chrome-test-runner

Chrome Test Runner Plugin

A Claude Code plugin that turns Claude into a QA testing agent. It drives your real Chrome browser, executes test scenarios, and produces unified diagnostic logs.

No custom MCP server needed — this plugin uses the Claude in Chrome extension that's already built into Claude Code.

What it produces

A unified markdown log that interleaves three layers for every test step:

• 🧠 Reasoning — what Claude is thinking, what it expects, what it's watching for
• 🎯 Actions — the browser interactions (click, type, navigate, scroll)
• 📡 Telemetry — network requests, console logs, JS errors, screenshots

Plus a bug report and summary statistics.

Setup

Prerequisites

1. Claude Code installed and up to date
2. Claude in Chrome extension installed in Chrome

Install the plugin

From local directory (development):

claude --plugin-dir /path/to/chrome-test-runner-plugin-v2

From GitHub:

# Inside Claude Code:
/plugin add github:your-username/chrome-test-runner-plugin

Start Claude Code with Chrome

claude --chrome

Or enable it permanently: run /chrome inside Claude Code and select "Enabled by default".

Usage

/chrome-test-runner:test https://myapp.vercel.app — stress test the chat, send rapid messages, create child chats

Or just describe what you want in natural language — Claude will auto-delegate to the test-runner agent:

Can you click around my app at localhost:3000 and find bugs?
Send several messages, try concurrent operations, and capture everything.

What's in the box

chrome-test-runner-plugin-v2/
├── .claude-plugin/
│   └── plugin.json          # Plugin metadata
├── agents/
│   └── test-runner.md       # The QA test execution agent
├── skills/
│   └── chrome-testing/
│       └── SKILL.md         # Testing knowledge base (auto-loaded)
├── commands/
│   └── test.md              # /chrome-test-runner:test slash command
└── README.md

Component	Purpose
test-runner agent	Plans test scenarios, executes them with the flush→act→capture cycle, writes reasoning, flags bugs, compiles the unified log
chrome-testing skill	Auto-loaded knowledge about bug patterns, telemetry tools, console message triage, JS inspection snippets
test command	/chrome-test-runner:test shortcut to trigger the agent

How it works

The test-runner agent uses Claude in Chrome's built-in tools:

Tool	Purpose
navigate	Go to URLs
computer	Click, type, scroll, screenshot
find	Locate elements by description
form_input	Fill form fields
read_network_requests	Capture HTTP traffic
read_console_messages	Capture console output
javascript_tool	Run JS for deep inspection
gif_creator	Record interaction GIFs

The key pattern is flush → act → capture: before each action, clear the telemetry buffers; after each action, read fresh telemetry. This gives you clean per-step data instead of one giant dump at the end.

Example output

See EXAMPLE-LOG.md for a full sample of what the unified log looks like.