mirror of
https://github.com/stablyai/orca.git
synced 2026-10-07 16:02:29 +00:00
Brennan Benson
e9ed75a5de
* fix(native-chat): mark a chat whose start failed on its tab and workspace row A chat whose start the host refused said so only inside its own tab. Its row now carries the failed verdict the host publishes for a refused start, so the tab and the workspace row show it from elsewhere until a retry starts it or the tab closes. The + menu no longer shows a failed start as still starting, so launching the agent again retries it. * fix(native-chat): date the failed-start mark by the failure; keep the + menu change for a separate PR The failed-start mark on a chat's tab and workspace row is now dated by when the start failed, the way a failure the host reports is dated: it ages out on the same schedule, a restart does not bring an old mark back, and a failure the user already saw stays read. The failure time is saved with the launch record so a restart keeps it; records saved without it fall back to the tab's creation time. The + menu change (letting a failed agent be picked again) moves to its own PR: its correct form makes a new chat instead of retrying the failed one, which changes how launches join. * test: name the local host on the saved failed-launch record * fix(native-chat): re-check an unconfirmed chat start when its host is reachable again A chat create whose reply was lost, or that a reload interrupted, stayed unknown until the user pressed Retry in its tab. Each host subscription opens with one full inventory; on it, every launch that host owns whose outcome is unknown (including ones restored from the saved launch record) is re-checked through the existing Retry path, so it settles to started or failed. One still unknown waits for the host's next return; cancelled launches are skipped.