Browser Safety

Rules that apply to every skill in this toolkit that drives a browser. They exist because each one corresponds to something that has actually gone wrong in a real run, not to a hypothetical risk.

This skill does not drive the browser itself. The mechanism is @playwright/cli (install its official skill with playwright-cli install --skills) or the Playwright MCP server's browser_* tools.

Credentials

Never fill a real credential as a command argument or tool parameter. playwright-cli fill <ref> "$SECRET" prints the generated code with the secret resolved — await page.getByRole('textbox', { name: 'Password' }).fill('theActualPassword') — so the value lands in the transcript even though the shell expanded it and nobody typed it. MCP tools are no better: the literal has to appear in the tool call itself. This is verified behaviour, not a theoretical risk.