playwriter

name: playwriter description: Control the user's currently open Chrome tab through the Playwriter CLI (no new browser launch). Use when you need to inspect live UI state, run scripted browser actions, capture console output, or reproduce frontend issues directly in the user's tab.

Use this skill to drive the user's active Chrome tab via Playwriter.

Full documentation is available here: https://playwriter.dev/

Quick Start

Ensure the Playwriter extension is enabled (green) on the target tab.
Ensure CLI is available:

playwriter --version || npx -y playwriter --version

Create/attach a session:

playwriter session new

Run commands against that session:

playwriter -s 1 -e "console.log(await page.url())"

Core Workflow

Confirm connection and correct tab:

playwriter -s <session> -e "console.log(await page.url()); console.log(await page.title());"

Collect page structure when needed:

playwriter -s <session> -e "console.log(await accessibilitySnapshot({ page }))"

Execute targeted actions (click/type/hover/fetch/evaluate).
Pull logs and structured state via page.evaluate.
Summarize findings with exact IDs, timestamps, and observed state transitions.

Useful Commands

Get list rows/options from current app UI:

playwriter -s <session> -e "const rows = await page.getByRole('option').all(); console.log(rows.length);"

Read popup/hover content:

playwriter -s <session> -e "const row = page.getByRole('option').nth(0); await row.hover(); await page.waitForTimeout(700); console.log(await page.locator('[data-side]').first().innerText());"

Run arbitrary in-page debug code:

playwriter -s <session> -e "const out = await page.evaluate(() => ({ href: location.href })); console.log(out);"

Troubleshooting

If the session attaches to the wrong tab, click the extension icon on the intended tab and re-run playwriter session new.
If playwriter command is missing, use npx -y playwriter ... or install globally.
If execution errors suggest stale connection, create a fresh session.

Guardrails

Prefer read-only inspection unless the task requires mutation.
Announce destructive UI actions before running them.
When capturing logs, redact sensitive tokens/user data in summaries.