Baonat
← All skills
Testing · ClaudeVerified

Author: Anthropic

Claude Webapp Testing — Playwright against your local app

Code-side Playwright tests for a local webapp (selectors, screenshots, console) — not UI click-path computer-use QA. Use when you want scripts you can keep in the repo.

What it does
Steers Claude to recon the running app, discover selectors, and write/run Playwright flows (with helpers for server lifecycle) so failures are reproducible scripts, not one-off agent clicks.
Best for
Frontend engineers and agent users who already run a local dev server and want automated browser checks they can commit.
How to install / use
Enable webapp-testing from anthropics/skills (example-skills). Start the app, give base URL + happy path, ask for Playwright scripts with wait-for-networkidle and explicit assertions.

Prefer this when the test should live as Playwright code — use computer-use QA for ad-hoc UI click smoke instead.

Copy this

  • Contrast: computer-use QA = live UI clicks; this skill = Playwright scripts in-repo
  • Always wait for networkidle on dynamic apps before asserting
  • Prefer role/text selectors over brittle deep CSS chains

When to use

Use Anthropic’s webapp-testing skill when you need Playwright automation against a local webapp — scripts, screenshots, console inspection, and assertions you can keep in the repo.

Contrast with Computer-use QA: that card is UI click-path smoke via Anthropic computer use (agent drives a real browser/desktop session). This card is code-side: Claude helps write and run Playwright against your local server. Pick computer-use for a short ad-hoc staging walk; pick webapp-testing when the outcome should be durable test code.

Good fits:

  • Local npm run dev / static HTML smoke before a PR.
  • Reproducing a UI bug with a minimal Playwright script + screenshot.
  • Capturing console errors while exercising a flow.
  • Teaching an agent to recon DOM selectors, then act (recon → action).

Skip it when you have no runnable local app, when you only need API/unit tests, or when you specifically want computer-use click-path QA on staging without writing Playwright.

Why

Agents that only “look and click” once do not leave an artifact your CI can rerun. Anthropic’s public webapp-testing skill (in anthropics/skills) biases toward Playwright scripts, server helpers, and wait strategies so checks are repeatable.

Baonat marks this verified. Free to clone from GitHub (example-skills / Apache-2.0). Host Claude usage may still follow your plan. It complements — does not replace — claude-computer-use-qa.

Install

  1. Get the skill from skills/webapp-testing, or install Anthropic’s example-skills plugin where available.
  2. Start the app locally (or use the skill’s server helper pattern) and note base URL + port.
  3. Describe the flow in ≤8 steps with visible copy to assert.
  4. Ask for Playwright with networkidle waits, role/text selectors, and screenshot on failure.
  5. Run once, fix selectors, then commit the script.
  6. Do not point at production customer data.

Copy this

Kickoff prompt:

Attach the webapp-testing skill (Playwright — NOT computer-use click QA).
Local URL: [http://localhost:PORT]. Start command: [npm run dev].
Flow (max 8 steps):
1) …
Assert visible copy / roles at each step. Wait for networkidle on dynamic pages.
Deliver: Playwright script + how to run + screenshot-on-failure notes.
Prefer getByRole / getByText. No production credentials or real PII.

Concrete scenarios

  1. PR smoke. Login → create item → expect success toast; script committed under tests/e2e.
  2. Bug repro. Capture screenshot + console when a modal fails to open.
  3. Static page check. Read HTML, target selectors, run Playwright without a long-lived server.

Pitfalls / when NOT to use

  • Wrong tool. Need ad-hoc staging clicks with screenshots and no repo script? Use computer-use QA.
  • No waits. Dynamic apps flake without networkidle (or explicit ready checks).
  • Prod secrets. Keep tests on local/staging fixtures only.

FAQ

Q: How is this different from claude-computer-use-qa?
A: Computer-use QA = agent UI click-path. Webapp-testing = Playwright code-side local testing you can keep and rerun.

Q: Official Anthropic skill?
A: Yes — webapp-testing under anthropics/skills / example-skills.

Q: CI?
A: Treat the generated Playwright as a starting point; wire it into your own CI runner when stable.

Source attribution