You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug] SSH worktree cards die after every app restart (data loss, 100% repro) #18173
Setup: Windows client, direct-SSH target on Debian 12, Claude Code as agent.
If I close or auto-update Orca while a Claude session is running in an SSH worktree, that worktree's card is dead on the next launch: no status dot, no tabs, and clicking the card does nothing at all. The relay log on the host shows the client reconnecting fine but zero PTY attach traffic when I click, so the click dies client-side. The agent itself keeps running on the host the whole time, only the client tab state is lost. Cards without a running agent restore fine.
Repro: create SSH worktree > start Claude Code > send one message > quit and relaunch Orca. Happens every time, I reproduced it 3x including with a fresh worktree in a different base path.
This looks exactly like GitHub issue #12721 (SSH tab state written to a partition nothing reads, then synced back as deletion). It's confirmed there by three users on two platforms and the fix PR #12722 has been sitting since Aug 19.
Since sessions surviving a restart is kind of the point of the SSH setup, would be great if #12722 could get a review. Happy to test a build or send logs.
Re-audited after the Discord auto-archive recovery. Closing as fixed on current main: Merged #19572 fixes the exact SSH owning-partition split described here and explicitly supersedes #12722. Boot hydration now reads the owning SSH partition, adopts stranded rows, and preserves tabs across restart; behavioural round-trip and cold-activation coverage is present.
Activity
orca-discord-issues commentedon Sep 2, 2026
via DiscordMaaMcan you try the latest Orca version 1.4.195? It should fixed itAmethystLiang commentedon Sep 2, 2026
thanks for reporting. looking into it
nwparker commentedon Oct 7, 2026
Re-audited after the Discord auto-archive recovery. Closing as fixed on current
main: Merged #19572 fixes the exact SSH owning-partition split described here and explicitly supersedes #12722. Boot hydration now reads the owning SSH partition, adopts stranded rows, and preserves tabs across restart; behavioural round-trip and cold-activation coverage is present.Evidence: #19572, https://github.com/stablyai/orca/blob/acea59c6d874dbb2ffdce98f4e34a2e649b82a83/src/renderer/src/lib/workspace-session-host-hydration.ts, https://github.com/stablyai/orca/blob/acea59c6d874dbb2ffdce98f4e34a2e649b82a83/src/shared/workspace-session-stranded-partition-adoption.ts.
If the same problem persists on a build containing this change, please reply with the version and reproduction steps so we can reopen it.