Playwright CLI Skill: make browser checks reviewable

Give the agent one browser task

The official playwright-cli Skill describes a CLI workflow built around page snapshots, element references, actions and screenshots. A useful starting task is verifying that an add-item flow changes the visible list in your own test application. Name the expected result before asking the agent to interact with the page.

The project documents Node.js 18 or newer, installation of @playwright/cli and a separate install --skills step. Sessions are headless by default; a headed option lets you observe the browser. Treat saved authentication state as sensitive material rather than attaching it to a public test report.

Preserve the observed limitation

In issue 489, cyppe reports that attaching CLI 0.1.22 with Playwright 1.64.1 to an existing Chromium 154 browser on Linux can hang when one of its 28 tabs has a partially initialized frame. The report says later commands queue behind the blocked title lookup. It was open when checked on October 28, 2026. This is a specific existing-browser attachment report, not proof that every new session fails, and we have not verified an upstream fix.

Boloop recommends beginning with a separate test session and keeping the input, expected result and visible outcome together. We have not installed or run this external Skill for this recommendation.

Connect browser evidence to acceptance

A screenshot documents a state; it does not by itself establish that all requirements, permissions and recovery paths work. Use the AI-generated code acceptance checklist to record what you checked and what remains unverified.