browser-automation

Use the browser REPL, which provides a persistent Playwright-compatible page object connected to the user's real browser with cookies and logged-in accounts. Attach to an open tab with attachActiveBrowserTab() or open a new one with openTab(), then drive interactions through snapshot ref IDs.

Browser exec spawns an autonomous agent session for whole-task delegation across accounts and history, while the REPL gives direct low-level Playwright control for deterministic steps, screenshots, and downloads. Choose exec for delegation and REPL for precise verification.

Take a snapshot with snapshot(page, { interactive: true }) to get an accessibility tree with unique ref IDs like e12, then pass those refs to page.locator(). Refs are invalidated after each action, so always take a fresh snapshot before the next interaction.

Yes, use page.waitForEvent('download') before clicking the export button, then verify the result with download.path() and optionally save a copy with download.saveAs(). The fetch() helper only works for same-origin trusted direct-download URLs.

Each action invalidates previous snapshot ref IDs, so stale refs cause failures. Take a new snapshot after every action and treat the action as unconfirmed until a fresh snapshot shows the expected state.