Problem
Agent Mode follow-up turns can finish and write the final answer into the agent transcript without giving the user a clear completion cue. In a voice-first flow, this makes it feel like OpenClicky silently stopped responding.
Example from logs
During the App Store Connect W-9 check on 2026-06-12:
- 22:22:02Z: User asked
Did I fill this form out correctly as just an individual person?
- 22:22:02Z: OpenClicky logged
openclicky.request.completed for route=agent.followup, status=success, source=agent_overlay_open.
- 22:22:29Z: The agent produced the final answer:
Yes — for an individual person, this W-9 looks generally filled out correctly...
- 22:22:29Z: OpenClicky logged
openclicky.agent_completion.speech_budget events for that answer.
- No matching
OpenClicky task done desktop notification appears for that W-9 follow-up turn in the surrounding log window.
- The user then asked why OpenClicky did not talk or at least say it was done.
Actual behavior
The final answer is present in the agent transcript, but the user gets no obvious spoken or visual completion cue for the follow-up turn.
Expected behavior
When an Agent Mode follow-up turn completes, OpenClicky should provide at least one reliable completion cue:
- a desktop/HUD
OpenClicky task done notification, or
- a short spoken task-finished summary when voice mode initiated or steered the task, or
- an explicit HUD state transition from running to done.
Suggested fix
- Treat agent follow-up completion as a first-class lifecycle event, not only initial
agent.start completion.
- If
openclicky.agent_completion.speech_budget is generated for a final answer, ensure the corresponding done cue/notification path is triggered.
- Avoid marking the queue/submission success at 22:22:02Z as the actual task completion; the user-visible completion should occur when the final assistant message lands at 22:22:29Z.
- Add a regression test for
agent.followup turns that produce a final answer and verify a done notification or spoken summary is emitted.
Problem
Agent Mode follow-up turns can finish and write the final answer into the agent transcript without giving the user a clear completion cue. In a voice-first flow, this makes it feel like OpenClicky silently stopped responding.
Example from logs
During the App Store Connect W-9 check on 2026-06-12:
Did I fill this form out correctly as just an individual person?openclicky.request.completedforroute=agent.followup,status=success,source=agent_overlay_open.Yes — for an individual person, this W-9 looks generally filled out correctly...openclicky.agent_completion.speech_budgetevents for that answer.OpenClicky task donedesktop notification appears for that W-9 follow-up turn in the surrounding log window.Actual behavior
The final answer is present in the agent transcript, but the user gets no obvious spoken or visual completion cue for the follow-up turn.
Expected behavior
When an Agent Mode follow-up turn completes, OpenClicky should provide at least one reliable completion cue:
OpenClicky task donenotification, orSuggested fix
agent.startcompletion.openclicky.agent_completion.speech_budgetis generated for a final answer, ensure the corresponding done cue/notification path is triggered.agent.followupturns that produce a final answer and verify a done notification or spoken summary is emitted.