E2E 0.22 changes post-login screenshots and adds hosted test targets

E2E 0.22 and its accompanying engine releases are covered together. The most consequential change is what the model can see after a secret is entered.

Review screenshot exposure before upgrading

Filling a secret no longer disables model screenshots. Secure input fields remain masked and text output still redacts secret values, but a secret displayed in a plain input or elsewhere on the page can appear in the image sent to the model provider. The release explicitly recommends test credentials. A show-password control or an application echoing a value therefore needs attention; text redaction is not equivalent to image redaction.

This restores vision-based assertions and pixel actions after sign-in, including tests that restore sessions. It also removes the previous pixel-taint interface, so integrations using that interface need a migration review.

More ways to drive and inspect a test

The release adds point-to-point dragging and selective recording through start/stop recording calls. The BrowserStack engine leases real Android and iOS devices through App Automate, while the ACP adapter adds an agent integration. Related web and mobile packages are part of the same release group, not separate product launches.

Mobile target selection avoids booted devices held by another agent-device session. Cleanup failures now affect the exit result and appear in the report, rather than leaving a misleading green summary. These are documented behavior changes; no reliability or speed benchmark was performed. Hosted-device accounts and model-provider costs are separate from the framework.

See the e2e hub.