Playwright

Playwright

Overview

Use this skill for Playwright work across three common outcomes:

• checked-in automated browser tests
• checked-in browser automation scripts
• repeatable Codex browser playbooks

In this skill, a "playbook" means a reusable browser workflow for Codex. Implement playbooks in the form that best matches the task:

• Playwright Test when the workflow should become regression coverage in the repo
• Playwright library scripts when the workflow is a utility or one-off browser automation program
• Playwright CLI or Playwright Model Context Protocol server workflows when the task is agent-driven browser control and the integration is already available

Treat the official Playwright documentation as the canonical and most up-to-date source. Use this skill as a working guide, not as a substitute for the official docs. If command syntax, configuration, API behavior, language support, or tool availability is uncertain, re-open the official docs before finalizing.

• Playwright homepage
• Writing tests
• Best Practices
• Locators
• Assertions
• Isolation
• Authentication
• Projects
• Page object models
• Test generator
• Trace viewer
• Auto-waiting and actionability
• Coding agents CLI
• Model Context Protocol server

Workflow

1. Identify which Playwright mode actually fits the task:

• checked-in tests
• checked-in scripts
• agent playbook

1. Inspect the repo before drafting:

• detect language and package manager
• detect existing Playwright usage, config, fixtures, auth setup, projects, reporters, and local dev-server wiring
• preserve repo conventions unless they are clearly brittle or incorrect

1. Choose the existing repo binding first:

• stay in TypeScript or JavaScript repos that already use @playwright/test
• stay in Python, Java, or .NET repos that already use those Playwright bindings
• do not introduce a different binding unless the user explicitly wants that

1. Open the reference file that matches the task:

• [references/playwright-mode-selection.md](references/playwright-mode-selection.md) for deciding between Playwright Test, library scripts, Playwright CLI, or Model Context Protocol server workflows
• [references/playwright-authoring.md](references/playwright-authoring.md) for locators, assertions, actionability, code generation, and page object guidance
• [references/playwright-architecture-and-debugging.md](references/playwright-architecture-and-debugging.md) for fixtures, projects, authentication, local servers, traces, and suite structure

1. Draft the smallest change that solves the task without fighting the repo.
2. Re-open the official Playwright docs if any behavior is uncertain. Do not guess.
3. Validate against the review checklist before presenting the result as finished.

Codex Execution Notes

• Prefer the repo's existing Playwright language binding and file layout over personal preference.
• Prefer the repo's existing package manager and command style over generic npm examples.
• Prefer checked-in tests for durable coverage and checked-in scripts for utilities. Use agent-only workflows when the task is inherently transient.
• If Playwright CLI or the Model Context Protocol server is not configured in the environment, do not pretend it is. Fall back to repo-local Playwright usage.
• Cite the official Playwright docs in user-facing answers when the user wants rule sources, latest guidance, or explanation of a Playwright behavior.
• Treat generated code from codegen as a starting point that usually needs refactoring.

Core Rules

• Use resilient locators first. Prioritize user-facing attributes and explicit contracts.
• Prefer getByRole, getByLabel, getByPlaceholder, getByText, getByAltText, getByTitle, and getByTestId over brittle selectors.
• Use Playwright's web-first assertions such as await expect(locator).toBeVisible().
• Rely on auto-waiting and actionability checks instead of manual sleeps.
• Keep tests isolated. Assume each test should stand alone unless the suite deliberately uses a different pattern.
• Use authentication state intentionally and keep it out of version control.
• Use projects for browser, device, environment, or authenticated-state matrices when that matrix is part of the requirement.
• Use page objects only when repetition and maintenance pressure justify them.
• Use trace viewer to debug failures rather than guessing from screenshots alone.
• Use local web-server config when tests depend on a local app startup sequence.

Do Not

• Do not assume this skill is more current than the official Playwright docs.
• Do not guess on Playwright API behavior, CLI flags, configuration names, or language-binding differences when the official docs can answer it.
• Do not introduce Playwright into a repo without first checking whether the repo already has a browser-testing strategy and conventions.
• Do not switch a repo from one Playwright binding to another unless the user explicitly wants that migration.
• Do not default to long CSS or XPath chains when role, label, text, or test id locators exist.
• Do not use first(), last(), or nth() as the first answer to locator ambiguity.
• Do not rely on page.waitForTimeout() or frame.waitForTimeout() in production test code. Those are for debugging only.
• Do not rely on networkidle readiness checks for test correctness when web assertions can express readiness directly.
• Do not write manual assertions such as expect(await locator.isVisible()).toBe(true) when a web-first assertion exists.
• Do not bypass actionability with force unless there is a concrete, justified reason.
• Do not commit authentication state files.
• Do not test third-party services you do not control when mocking or routing the dependency is the better design.
• Do not accept raw codegen output without review and cleanup.
• Do not over-abstract early with fixtures or page objects before the suite actually needs them.
• Do not use stale ElementHandle-style flows for ordinary user actions and assertions when locators are the correct abstraction.

Review Checklist

• Confirm the chosen Playwright mode matches the task.
• Confirm the change follows the repo's existing language binding, test runner, and package-manager conventions.
• Confirm locators are resilient and user-facing where possible.
• Confirm assertions are web-first and retry-aware.
• Confirm there are no unnecessary manual waits, brittle selectors, or unjustified forced actions.
• Confirm authentication state, if used, is stored safely and ignored by version control.
• Confirm fixtures, projects, and page objects are justified by repetition, matrix coverage, or suite scale.
• Confirm debugging guidance uses traces, reports, or Playwright-native tools instead of guesswork.
• Confirm any uncertain API or configuration rule against the official Playwright docs before presenting the result as final.

References

• Read [references/playwright-mode-selection.md](references/playwright-mode-selection.md) for deciding whether the task should become a test, script, or agent playbook.
• Read [references/playwright-authoring.md](references/playwright-authoring.md) for locators, assertions, actionability, codegen cleanup, and authoring rules.
• Read [references/playwright-architecture-and-debugging.md](references/playwright-architecture-and-debugging.md) for fixtures, projects, authentication, local web servers, traces, and suite maintenance.
• Use the official Playwright docs as the canonical source:
• Playwright
• Writing tests
• Best Practices
• Locators
• Assertions
• Isolation
• Authentication
• Projects
• Page object models
• Test generator
• Trace viewer
• Auto-waiting
• Coding agents CLI
• Model Context Protocol server

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: andrey-learning-machines
• Source: andrey-learning-machines/swe-harness
• License: MIT

Install and usage instructions live in the source repository linked above.