Web App Testing skill: Skill for Claude Code, OpenAI Codex & more

Works with

Built on the open Agent Skills standard, so it works in every app that supports Agent Skills.

Amp

Zed

Devin Desktop (formerly Windsurf)

About Web App Testing skill

What it does

Web App Testing is a skill from Anthropic's official repository that teaches the agent to check a local web application with Playwright. The agent does not simply claim a UI change works. It writes a short Python Playwright script, runs it against the app in a headless Chromium browser, and then reads the results: screenshots, the rendered DOM and console output. This lets the agent confirm its own frontend work and debug layout or behaviour problems without an extra MCP server.

What is inside

The SKILL.md gives the agent a decision tree. For a static HTML file, it reads the file to find selectors. For a dynamic app, it starts the server or uses the one already running, waits for the page to settle, looks at the page first, and then acts.

The skill ships a helper script, scripts/with_server.py, which starts one or more dev servers (for example a backend and a frontend), waits for their ports and then runs your automation. It also includes three example scripts: one finds buttons, links and inputs; one automates a local HTML file; one captures console logs.

Works with

It is part of the example-skills plugin in the Anthropic marketplace for Claude Code. As a standard SKILL.md folder it can also be copied into other agents that support Agent Skills, provided they can run Python.

How to install

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

You also need Python with Playwright and its Chromium browser installed. See the Playwright documentation for setup.

Maintenance and safety

The skill is maintained in the main Anthropic skills repository and is licensed Apache 2.0. It runs real code on your machine: the helper starts whatever server command you give it, and the generated scripts drive a browser against your app. Review the commands before you approve them.

Who should use it

Anyone building web apps with an agent who wants it to check its UI work before calling the task done. If you prefer interactive browser control through tools rather than scripts, a Playwright MCP server is the alternative.

testing playwright