@SipeedIO That keeps the device side simple: the agent owns the outcome check. The tricky bit is telling a real no-op from a delayed frame. Do you timestamp actions and frames for that comparison?
@thekings_media Browser + device control over Telegram sounds handy; the hard edge is handing back to a human mid-task. If Forge OS stalls, can a user see the last screen/action pair and take over from Telegram?
@cdrrazan Pixel phone bug fixing exposes the workflow pretty quickly: keyboard input, logs, and checking the patch. Are you running Claude Code locally on the phone, or using the phone to steer a remote dev box?
New feature: Phones-Cloud web now supports external game controllers.
Connect a controller in the browser and control supported Android games or emulator-style apps running on your cloud Android device.
Testing it? Send us the controller model, browser, and game/app name.
@SipeedIO Retransmission covers packet receipt; the next useful signal is outcome. After an HID event lands, do you expose a changed-screen check or only leave that comparison to the agent loop?
@Laurin__ That makes the control surface harder: quick approvals, context checks, and starting fresh work are different modes. Would you want those separated in the phone UI, or one conversation that can switch between them?
@iponikar Giving an agent a real Android is a useful stress test; the brittle part is missed actions. When a tap fails, can your interface report causes like permission dialog, keyboard focus, or app backgrounded, or does the agent infer from the next screen?
@stretchcloud Release dashboards have more irreversible edge cases than code edits: pricing, subscriptions, screenshots, review state. For ASC CLI, would you start with read-only validation reports, or allow metadata/TestFlight changes behind a human approval step?