Skales - AI Agent & Tasks

This release fixes a crash that could stop Skales from opening at all on iOS 26. Also fixed: A long answer is now read to the end. The safety timer that watches read-aloud used to cap every reading at two minutes, so anything longer than about 1,300 characters was cut off mid sentence. The timer now watches for a reading that has actually stalled, so length is never punished and a genuine hang still recovers. Every agent in the chat picker is reachable. The list under the height cap could grow past the screen with no way to scroll, and "Manage" was the last row, so the way out disappeared with it. The rows scroll now, the title stays put, and Manage is a fixed footer. The goal sheets in Chat and Code had the same flaw and got the same fix. A start that fails names itself. The app now carries a safety net above every screen, the preference store survives a failed start with a named fallback instead of dying silently, and the entry decision can no longer leave the app on a black screen. The last start's steps travel with the crash report, so the next mystery is a measured one. Small on device models are no longer handed instructions they cannot follow. Very small local models were sent a protocol built for larger ones and could repeat themselves until stopped by hand. Models that small no longer receive it, and a reply that starts looping stops with a reason on screen. Larger models are unchanged. Typing in Flow stays visible. The model, style and video pickers now move out from under the keyboard, the way every other input surface in the app does. Read aloud follows the language of the text, not the language of the interface, and a voice you chose yourself always speaks, even when the text is in another language, because that was your choice and not an accident. New: a female Spanish voice on the device. Sharvard joins the on device voices, downloadable from the same list as the others.