Auto-collapse thinking and tool activity after turns - #83
Conversation
|
Thanks for this @jagadhis The auto-collapse looks good overall, but expanding the summary changes the transcript order. For example, tool A → progress message → tool B becomes progress message → tool A → tool B. We need to preserve the original activity/prose ordering when the summary is expanded? |
|
Fixed in 55fbd40. Expanded activity now renders each original activity group at its existing TurnItem position, so tool A → progress message → tool B keeps that order. The compact summary remains immediately above the final answer and still toggles the whole trail. I also reran npm run check successfully (1,249 web tests and 198 Rust tests). |
|
I appriacete your efforts @jagadhis I tested this locally, and the collapsing causes too much layout and scroll shifting. Content jumps when tools are changing, when the turn settles, and when the summary is expanded. I think this needs to be handled in a different way. I would imagine, and that was the original idea but i haven't gotten the time to work on it, is to comabine as much of the tool calls withouth loosing context while streaming and once settleed to get the final answer. In its current form, the movement is too disruptive to merge |
2 participants
What changed
Finished turns now fold all thinking and tool activity into one expandable summary above the final response. A persisted Settings → General toggle keeps the behavior optional, while live turns remain expanded and failed or rejected calls stay visible in the summary count.
Why
Long agent runs currently leave several process phases above the final answer, making completed conversations difficult to scan. This implements the compact settled state requested in #79 without changing the live monitoring experience.
UI
No screenshot attached. The change uses the existing transcript phase styling and adds the
Auto-collapse activitytoggle under Settings → General.Checklist
npm run checkCloses #79