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
With Auto-generate tab titles on, split a terminal tab and run an agent in each pane. Every agent row for that tab shows the title generated from one pane's prompt, so the other pane's agent is listed under a task that isn't its own. This affects the worktree-card rows, the dashboard/board cards and the Agents sidebar list.
Steps
Turn on Settings → Auto-generate tab titles.
Open a terminal tab and split it.
In the left pane, start an agent with "Fix the patient intake flow in the portal".
In the right pane, start an agent with "Design a Redis cache strategy for sessions".
Look at the two agent rows under the worktree card, then switch the sidebar to the Agents list.
Expected: each row is named after its own pane. Actual: both rows read "Fix the patient intake flow in the", in both views.
Why
generatedTitle is stored once per tab. applyGeneratedTabTitleUpdates (terminal-tab-title-batch.ts) derives it from the first prompt seen in any pane of the tab and writes it tab-wide.
The per-pane row resolvers use it for every pane of that tab: getAgentRowConversationName and getActivityThreadTaskTitle.
fix(sidebar): let an agent row show the provider's own session title #19936 closed the same leak for provider session titles ("a provider session title is now accepted only when both the agent and provider-session identities match, so sibling panes retain their own live titles"). generatedTitle is the remaining tab-wide name that leaks across panes.
Suggested fix
Record which pane produced the generated title when it is written, and let a row use the title only for that pane. This is the same scoping #19936 applies to provider titles.
A pane that named the tab before a later split keeps its title, and so does a lone agent next to a shell.
Titles saved before the change keep today's behavior.
Operating system
macOS
Orca version
v1.4.203 (also reproduced on main
caa465d1d)Details
With Auto-generate tab titles on, split a terminal tab and run an agent in each pane. Every agent row for that tab shows the title generated from one pane's prompt, so the other pane's agent is listed under a task that isn't its own. This affects the worktree-card rows, the dashboard/board cards and the Agents sidebar list.
Steps
Expected: each row is named after its own pane.
Actual: both rows read "Fix the patient intake flow in the", in both views.
Why
generatedTitleis stored once per tab.applyGeneratedTabTitleUpdates(terminal-tab-title-batch.ts) derives it from the first prompt seen in any pane of the tab and writes it tab-wide.getAgentRowConversationNameandgetActivityThreadTaskTitle.customTitleandquickCommandLabel, but not for a title derived from one pane's prompt.generatedTitleis the remaining tab-wide name that leaks across panes.Suggested fix
Record which pane produced the generated title when it is written, and let a row use the title only for that pane. This is the same scoping #19936 applies to provider titles.
I have a PR ready for this.