PlugWiki

Webapp Testing skill for Claude: test your local web app with Playwright

Anthropic's agent skill that makes Claude write and run Playwright scripts to test your local web app in a headless browser.

Installs
195K
on skills.sh
Stars
186.4K
on GitHub
Files
5
besides SKILL.md
Updated
Oct 2026
last repo commit

webapp-testing is Anthropic's agent skill for Claude. It makes Claude write Python Playwright scripts that open your local web app in a headless Chromium browser, click through it, take screenshots and read console logs. It also bundles one helper that starts your dev servers before a test and stops them after. You need Python and Playwright installed first, and the helper runs your server commands through the shell.

Good for

  • Developers who want Claude to click through their local app and report what broke
  • Anyone debugging a UI bug who wants screenshots and browser console logs
  • Teams that want quick smoke tests on a front end without writing the scripts by hand
  • Claude Code users who already run a dev server such as Vite or a Node backend

Install webapp-testing

Claude Code

  1. 1

    Copy this prompt.

    Install the webapp-testing skill in Claude Code. Run this command:

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

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Python 3.8 or newer.

The skill itself does not install Playwright or a browser. The two Playwright steps come from the Playwright for Python docs. If you skip them, Claude can usually install them when the first script fails with an import error.

From the project README. Source.

What's in the skill

When it loads

Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.

Sections in SKILL.md

  1. Decision Tree: Choosing Your Approach
  2. Example: Using with_server.py
  3. Reconnaissance-Then-Action Pattern
  4. Common Pitfall
  5. Best Practices
  6. Reference Files

Key facts

Author
Anthropic
License
Apache 2.0 (LICENSE.txt in the skill folder)
Installs on skills.sh
194,845 as of 2026-11-01
GitHub stars
186K for the whole anthropics/skills repo, 2026-11-02
What is inside
6 files: 1 helper script (with_server.py), 3 example scripts, LICENSE.txt and SKILL.md
Part of
The example-skills plugin in the anthropic-agent-skills marketplace
Needs
Python 3.8+, Playwright for Python, a Chromium download, your app's dev server command
Last change to this skill
2026-04-20 (license line); last content change 2025-12-01
Works in
Claude Code; Claude.ai on Free, Pro, Max, Ultra, Team and Enterprise per the help center (the README still says paid plans); Claude API as a custom skill

What it includes

  • Picks static HTML, running server or server not running

    Claude first checks whether your page is plain HTML or a dynamic app, and whether the server is already running. That choice decides which path it takes.

    Read · Decision tree

  • Looks at the rendered page before it clicks anything

    Claude opens the page, waits until it finishes loading, takes a screenshot or reads the page, picks selectors from what it sees, then runs the actions.

    Try: “Use webapp-testing to open localhost:5173, find the buttons on the page and click Sign in.”

    Runs code · Reconnaissance, then action

  • Starts your dev servers, runs the test, stops the servers

    Starts one or more servers, waits until their ports answer, runs your test command and shuts the servers down. Claude runs it with --help first and does not read the source.

    Try: “Use webapp-testing. Start the backend and frontend, then run the checkout test.”

    Runs code · scripts/with_server.py

  • Lists the buttons, links and inputs on a page

    Opens a page, prints the buttons, the first five links and the form fields, and saves a screenshot.

    Files · examples/element_discovery.py

  • Records browser console messages while you click through

    Loads the page at 1920x1080, clicks a link and saves every console message to a log file.

    Write · examples/console_logging.py

  • Fills and submits a local HTML file

    Opens an HTML file from disk, takes a screenshot, fills in the form, submits it and takes another screenshot.

    Write · examples/static_html_automation.py

  • Waits for the page scripts to finish before reading it

    The skill's one hard rule. On a dynamic app Claude must wait for the page to settle before reading it, or it sees an empty shell. The browser always runs headless.

    Read · Wait for networkidle

From the official docs

Permissions

Medium risk

Claude runs scripts on your machine, and the server helper passes your commands to the shell, so a careless command can cause real damage. Use it on projects you trust, with test data, and read the commands Claude proposes before you approve them.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • The server helper runs commands through a shell

    with_server.py passes your server command to the shell, and three issues report command injection (one closed as a duplicate). Only give it commands you wrote, and read what Claude proposes before you approve it. github.com

  • Dev servers can be left running or hang

    Open bug reports say it can leave node or vite processes running, hang on servers that print a lot, and report a pass when the port was already taken. None of the fix pull requests are merged, so check for stray processes and stop them yourself. github.com

  • Page content goes into the conversation

    Claude reads the page text, screenshots and console logs. A page with hostile text could try to steer Claude, so point it at your own app, not at unknown sites. skills.sh

  • No timeout on the test command

    A test that never ends keeps running until you stop it. Watch the first runs. github.com

  • The Claude Code plugin installs the whole example set

    The plugin route loads every skill in the example-skills set, not only this one. Use the skills command in the install box if you want just this skill. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details
Turning on read-onlyNo read-only mode. The scripts can click buttons and submit forms in your app, so use a dev database and test accounts.
Talks toPlaywright browser download at setup. The test scripts open localhost or file:// pages. Your AI provider receives page content, screenshots and console logs that Claude reads
Sign-in and keysNo account or key. Any login the app needs is whatever you type into the test.
Data retentionNo telemetry. The skill is plain files and scripts; screenshots stay where the script saves them.
Scopes

  • Claude writes Python scripts and runs them on your machine

  • with_server.py starts the server commands you give it and then runs your test command

  • Screenshots go to /tmp; two example scripts write to /mnt/user-data/outputs

Examples

  • Check that a running Vite app still works after a change

    Use the webapp-testing skill. My app is running on localhost:5173. Open it, click through the login and dashboard pages, and tell me what breaks.

    Claude opens the page, waits for it to finish loading, takes a screenshot, picks selectors from what rendered and runs the clicks.

  • Start a front end and back end together and test them

    Use webapp-testing. Start the backend in ./backend on port 3000 and the frontend in ./frontend on port 5173, then submit the signup form and check for errors.

    Claude uses the bundled server helper with two servers, runs the test and stops both servers afterward.

  • Collect browser console errors

    Use webapp-testing to load http://localhost:5173, click Dashboard and save every console message to a log file.

    Claude adapts the console logging example and writes the messages to a file you name.

  • Try a static HTML form

    Use webapp-testing on ./signup.html. Fill in name and email, submit, and screenshot before and after.

    Claude reads the HTML for selectors and runs a script against the file, saving two screenshots.

Questions

What is the webapp-testing skill?

An agent skill from Anthropic, in the anthropics/skills repo, that tells Claude how to test a local web app with Playwright. Claude writes Python scripts, runs them in a headless Chromium browser and reports what it sees.

How do I install webapp-testing?

Use the install box at the top of this page. The skill does not install Playwright, so also run the two Playwright setup commands shown there, or let Claude do it when it hits an import error. Python 3.8 or newer is required.

Does webapp-testing work in Claude.ai?

Yes, as an example skill. Anthropic's help center lists skills on Free, Pro, Max, Ultra, Team and Enterprise plans, with code execution turned on; the repo README still says paid plans. Enable it under Customize > Skills. Anthropic does not say whether that sandbox can launch a browser or reach a server on your own machine, so Claude Code is the safer choice for local apps.

Does webapp-testing work in Codex or Cursor?

Anthropic documents it for Claude only. It is plain files, so other tools that read skill folders may load it, but nothing from Anthropic says so. GitHub's awesome-copilot has a different skill with the same name that uses Node.

Is webapp-testing safe?

Mostly, if you use it on your own projects. Claude runs scripts and a helper that passes server commands to the shell. On skills.sh, Gen Agent Trust Hub passes it, Socket warns at low risk and Snyk warns at medium risk, all dated Sep 14, 2026. No GitHub security advisories are listed.

Why does the script fail to save a file to /mnt/user-data/outputs?

That folder only exists in Claude.ai's sandbox. Two example scripts write there, so on your own machine change the path to a folder you have, such as /tmp, or ask Claude to do it.

Why did my test pass even though the app was down?

The helper checks that something is listening on the port, not that it is your app. If an old server already uses that port, the test runs against it. Stop the old process or pick a free port. A fix is open but not merged.

Why is a dev server still running after the test?

The helper stops the servers it started, but open reports say child processes such as vite can survive. Check for leftover node processes and stop them yourself.

webapp-testing vs Playwright MCP?

webapp-testing has Claude write and run one-off Python scripts, so there is no server to connect and the steps stay in files. The Playwright MCP server gives Claude browser tools it calls one step at a time. Pick the skill for repeatable test scripts and the MCP server for exploring a page live.

webapp-testing vs playwright-cli or agent-browser?

Both are command line browsers Claude drives step by step. playwright-cli and agent-browser suit exploring a site live. webapp-testing is better when you want a script you can keep and rerun, plus the helper that starts your dev servers.

Information

updated Nov 2, 2026

Repository
anthropics/skills
Folder
skills/webapp-testing

anthropics/skillsskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 28
  1. anthropics/skills README · github.com
  2. Using skills in Claude · support.claude.com
  3. skills.sh listing · skills.sh
  4. Playwright for Python · playwright.dev
  5. Claude Code skills · code.claude.com
  6. webapp-testing SKILL.md · github.com
  7. with_server.py · github.com
  8. anthropics/skills · github.com
  9. LICENSE.txt · github.com
  10. marketplace.json · github.com
  11. Folder history · github.com
  12. Playwright for Python · playwright.dev
  13. element_discovery.py · github.com
  14. console_logging.py · github.com
  15. static_html_automation.py · github.com
  16. Output path pull request · github.com
  17. Fix for false pass on occupied port (open) · github.com
  18. Fix for orphaned dev servers (open) · github.com
  19. Snyk audit: Warn, medium risk, Sep 14, 2026 · skills.sh
  20. awesome-copilot webapp-testing (different skill) · github.com
  21. Gen Agent Trust Hub audit: Pass, Sep 14, 2026 · skills.sh
  22. Socket audit: Warn, low risk, Sep 14, 2026 · skills.sh
  23. Command injection report (open) · github.com
  24. Command injection report (open) · github.com
  25. Command injection report (closed as duplicate) · github.com
  26. Fix for missing timeout (open) · github.com
  27. Plugin install loads all repo skills · github.com
  28. GitHub security advisories: none listed on 2026-11-02 · github.com

Updated Nov 2, 2026.