Claude Playwright Skills: Best Browser Automation and E2E Testing

A Claude Playwright skill gives your agent a repeatable way to open a browser, check what a page actually does and write tests that keep passing. The directory lists hundreds of skills under browser automation and e2e testing. This guide sorts ten of them by job, with the tool each uses, what you must install, the licence and the main trade-off.

Start with Anthropic's webapp-testing skill if you want a tested pattern for a local app. Pick a CLI-style skill if you want to drive a site step by step. Treat project-specific suites as templates for writing your own.

Which skill fits which job

How these skills were chosen

Candidates came from the two topic hubs, starting with the official webapp-testing skill and then skills that had a stated licence, a clear description and a pass or info result in the directory's automated safety check. Ranking mixes repository popularity, how many other owners copy a skill, SKILL.md quality and safety, as explained on the about page.

The editorial team did not run these skills. Facts come from each SKILL.md and repository. Where a skill depends on a tool, the requirement is as the skill states it.

Playwright-based skills for Claude Code

Web application testing

Web application testing is an Anthropic skill under Apache-2.0 in the directory, and its frontmatter points to a LICENSE.txt for full terms. It gives a decision tree: read static HTML directly, or for dynamic apps start the server and then inspect before acting. A helper, with_server.py, starts one or several servers and runs your script. Scripts launch Chromium headless and wait for the network to go idle before inspecting the page.

• Best for: verifying frontend behavior, screenshots and console logs against a local app.
• Requirements: Python, Playwright and an installed browser.
• Trade-off: it produces one-off automation scripts, not a maintained test suite for CI.

Playwright CLI

Playwright CLI wraps the playwright-cli tool for navigation, forms, snapshots, screenshots, tracing and tabs. The agent opens a page, takes a snapshot, uses the element references it returns and takes a new snapshot after the page changes. The licence is Apache-2.0.

• Best for: step-by-step exploration and UI debugging from the terminal.
• Requirements: Node and npm; Playwright installs browsers.
• Trade-off: it prefers CLI commands over test specs, and its setup example uses a Codex home path, so adjust it for Claude Code.

Dev Browser

Dev Browser lets the agent write small scripts against Playwright's page API while pages stay open on a local server between runs. It starts a fresh Chromium by default and has an extension mode that connects to your own Chrome. It is Apache-2.0 and was updated recently.

• Best for: multi-step tasks where each script builds on the last.
• Requirements: Node or Bun with tsx, and the skill's server script.
• Trade-off: extension mode touches your real browser, so use it deliberately.

Skills that write and maintain e2e tests

These three belong to specific repositories. Most readers should not install them as-is, but each shows how to turn a team's testing rules into agent instructions.

Write and verify Playwright tests

This Appsmith skill writes a Playwright test from a prompt, runs it against a live deployment and retries with fixes up to three times, treating failures as either a test bug or a product bug. It requires an Appsmith deployment URL, credentials in an env file and Chromium. The directory's check reports an info-level note for it, which can mean an env-file mention or broad shell access, so read the file. Licence: Apache-2.0.

• Best for: a model of a write, run, triage, retry loop with a hard retry cap.
• Requirements: an Appsmith deployment and the repository's Playwright setup.
• Trade-off: it assumes deep knowledge of that project's test structure.

Handsontable Playwright e2e

Handsontable's skill sets five rules: page objects, data-testid selectors, web-first waits, isolated tests and shared fixtures. It also notes a gotcha with duplicated overlay elements and says to check that a port is free before running. GitHub reports its licence as NOASSERTION, so check the repository before reuse.

• Best for: teaching an agent your selector and wait conventions.
• Requirements: the Handsontable test setup.
• Trade-off: specific to one component library.

Cucumber and Playwright e2e

Dify's skill covers feature files, step definitions and locators in an e2e folder, keeps Cucumber and Playwright responsibilities separate and says to run the narrowest tagged scenarios first. Licence is also NOASSERTION in GitHub's data.

• Best for: BDD-style suites with a written governance file.
• Requirements: Cucumber, Playwright and the repository's AGENTS files.
• Trade-off: it points to files that only exist in that repository.

Browser automation skills that use other engines

Dogfood

Dogfood explores a web app through Vercel's agent-browser CLI, then writes a report with screenshots, repro videos and numbered steps for each issue. It targets five to ten well-documented issues, not an exhaustive list, and tells the agent to test as a user would without reading source. Apache-2.0.

• Best for: exploratory QA before a release.
• Requirements: the agent-browser CLI, which installs through npm, plus a Chrome download.
• Trade-off: it reports problems and does not fix them or create a repeatable test.

Chrome CDP browser control

Browser CDP connects to Chrome through a remote debugging port so the agent can reuse existing logins. It uses a persistent debug profile, and it must ask for consent before stopping running Chrome processes. MIT licensed.

• Best for: automating a site where you are already signed in.
• Requirements: Chrome, Node 20 or later and agent-browser.
• Trade-off: it acts as you, so use a dedicated profile and a low-risk account.

Skyvern

Skyvern browser automation selects among CLI commands: boolean checks, schema extraction, deterministic clicks, AI-guided actions, one-off tasks and reusable workflows. It says never to type passwords and to use stored credentials. It is licensed AGPL-3.0.

• Best for: repeatable multi-page workflows that you want cached.
• Requirements: the Skyvern CLI; the skill offers cloud sessions for public URLs and local sessions for localhost.
• Trade-off: its quick action mode reasons without screenshots, which can struggle on visually complex pages, and AGPL-3.0 has network-use terms.

Acceptance evidence

Acceptance evidence for deliveries is for proving a change works from a user's point of view, across CLI, web, desktop and iOS simulator surfaces. It excludes unit tests and lint from acceptance checks and treats published rounds as permanent records. Apache-2.0, but it is tied to LobeHub's tooling, so adapt it rather than install it.

Safety notes for browser skills

Browser skills act with real accounts and real pages, so read them more carefully than a formatting skill. Look for how each handles credentials, whether it attaches to your own profile, what it does with cookies and how it asks for help on CAPTCHAs. The skills above that touch logins say to avoid reading credentials and to hand verification steps back to you.

How to choose

• First install: webapp-testing. Our list of the best Claude Code skills covers other jobs.
• Terminal exploration: Playwright CLI or Dev Browser.
• Pre-release bug hunt: Dogfood.
• A real test suite: write your own skill, borrowing the retry cap from Appsmith's skill and the selector rules from Handsontable's.
• Logged-in sites: only with a separate profile and the stricter skills.

See the Claude Code page for install methods, and read each repository's licence before you rely on it.