Claude webapp-testing skill

A coding agent that says "the button works now" without opening the page is guessing. Anthropic's webapp-testing skill makes Claude check in a browser. It writes Python Playwright scripts against your local app, runs them headless, and reports what it saw. It is one of Anthropic's open-source example skills, under the Apache 2.0 license. For background, read how Claude Skills work.

What the skill does

The SKILL.md gives Claude a decision tree:

• Static HTML: read the file to find selectors, then write the script.
• Dynamic app, server not running: use the bundled with_server.py helper, which manages the lifecycle of one or more servers (for example a backend and a frontend) and runs the test script once they are up.
• Dynamic app, server running: "reconnaissance, then action". Load the page, wait for network activity to settle, take a screenshot or inspect the DOM, find selectors from what actually rendered, and only then click and type.

The skill names one common mistake explicitly: inspecting a JavaScript app before it has finished loading. Bundled examples cover finding buttons, links and inputs, testing local HTML files, and capturing console logs. Claude is told to run the helper scripts with --help rather than reading their source, which keeps its context small.

When to use it

• Confirming a fix in the UI before you commit it.
• A quick smoke test of a sign-up flow or form on your dev server.
• Screenshots of a page at a given state for a bug report.

Pair it with the bundled /code-review skill before a pull request, or with Superpowers, which adds test-first habits. Our Claude Code prompts include requests for test plans, and our AI coding tools guide compares the agents.

How to install it

Claude Code:

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

Claude.ai: the anthropics/skills README says its example skills are already available on paid plans, but the skill is built for apps running on your own machine, which is the Claude Code setting.

Example request

Use the webapp-testing skill. Start the app with [command] on port [port], then check that [user action, e.g. adding an item to the cart] works: take a screenshot before and after, and show me any console errors.

Limits

• It needs Python and Playwright with Chromium installed where Claude runs.
• Scripts are written per request. It does not create a lasting test suite unless you ask Claude to save them as one.
• It targets local apps. Testing a live site means pointing scripts at a public URL, which you should do only on sites you own.
• Anthropic labels its example skills as demonstrations and advises testing before you rely on them.