Webapp Testing · Field report I — Information
Anthropic source · independent redistributionPython PlaywrightVerified 2026-11-03

Observe the interface. Verify the change.

See what your interface really does.

Anthropic's Webapp Testing Skill helps a coding assistant inspect local web apps, test user interactions, capture screenshots and collect browser logs with native Python Playwright.

Its working discipline is simple: look at the rendered page before touching it. The assistant waits for the app to settle, reads the real DOM or a screenshot, then acts with selectors that actually exist — and keeps the evidence.

Method
Native Python Playwright scripts, headless Chromium
Helper
with_server.py starts one or more servers, waits for their ports, runs your test, cleans up
Examples
Element discovery · static HTML via file:// · console logging
Verified
4 of 4 browser checks on a synthetic local fixture
Source
Anthropic · Apache-2.0 · ca45fc496a34

Exhibit A — the idea in eight seconds

Overview film
8-second capability overview. Illustrative animation; the test evidence is the filmstrip below.
Real Playwright capture before the action: review queue showing three items, two Ready and one Reviewed

Frame 01 · rendered3 items · 2 Ready

Real Playwright capture after the review action: Reviewed filter showing two items and the confirmation message

Frame 02 · after actionReviewed · 2 items

4 / 4 pass

Exhibit B — one small fixture, one real interaction. Filter → click → read back the visible result. Both frames were captured by Playwright in the verified run; empty page margin is trimmed here, and the uncropped captures appear in Benchmark and Showcase.

§1

Purpose

A browser, with a plan.

The Skill teaches the assistant to test a local web app the way a careful reviewer would — by observing the rendered interface first, then exercising the behavior that matters, then saving proof.

Discover first

Wait for the app, inspect its DOM or a screenshot, then choose selectors from the controls that actually exist.

Verify next

Exercise a user flow, inspect the resulting state and save a screenshot or console log for review.

§2

Scope

Where it fits — and where it stops.

Built for developers and reviewers checking a local web interface.

Fits best

  • Checking a click, form or filter in a local app.
  • Investigating unexpected UI state.
  • Capturing a local result for review.
  • Automating a static HTML file through a file:// URL.

Know the boundary

  • It is a testing guide and helper, not a hosted QA service.
  • A fixture smoke test does not prove every application works.
  • It does not replace security, load or accessibility audits.
  • Arbitrary websites and production systems were not tested.
§3

Inputs & outputs

Bring your local app. Leave with evidence.

You provide

  • A local URL — or server start commands and ports
  • The flow to test and the expected outcome
  • Access to any required test account

You receive

  • Task-specific Playwright automation
  • Observed assertions: what passed, what failed
  • Browser screenshots and logs, as produced by your assistant
§4

Package

What is in the folder.

Six original upstream files, byte-identical to the pinned commit, plus two packaging additions.

SKILL.mdDecision tree, reconnaissance-then-action pattern, pitfalls and best practices.
scripts/with_server.pySingle or multiple server lifecycle: start, wait for each port (30 s default timeout), run your command, stop.
examples/element_discovery.pyDiscovering buttons, links and inputs on a rendered page.
examples/static_html_automation.pyDriving a local HTML file through a file:// URL.
examples/console_logging.pyCapturing browser console messages during automation.
LICENSE.txtApache License 2.0, with upstream copyright retained.
PROVENANCE.mdPackaging addition: upstream repository, pinned commit and redistribution note.
UPSTREAM-THIRD-PARTY-NOTICES.mdPackaging addition: the upstream repository-wide notice inventory, kept verbatim. It does not mean all listed components are bundled here.
§5

Requirements

What you need to run it.

Host assistant
A coding assistant with filesystem and command access.
Runtime
Python, the Playwright package and a compatible installed browser. They are installed separately and are not bundled.
Your app
The application under test, its dependencies and any test credentials are yours to provide.
Keys & cost
No additional API key is needed for the local helper. Using your host assistant, its model, or an external application may still incur costs.
§6

Verification

What has been verified.

Passed · scoped

From a clean ZIP extraction, the original with_server.py started a local synthetic fixture, ran a Python Playwright test and cleaned up the server. Four browser checks passed, screenshots were captured and no page errors were observed.

One local run on 2026-11-03. Multiple-server mode, console-log capture, static file:// pages, other browsers and real applications were not part of this run.