Skip to content

[Feature]: Preserve reasoning/tool trace expanded state when switching threads #3733

Description

@lastobelus

Summary

Feature request: add an option to preserve each thread's open/collapsed reasoning/tool-trace state when switching away from a thread and back.

If showing only the terminal assistant message is intentional behavior for most users, preserving the expanded state would still make it possible to recover user-facing content that was intentionally or heuristically grouped into the trace.

Motivation

Today, T3 Code can collapse substantial assistant output behind the Worked for ... reasoning/tool trace. When I expand that trace, the missing content is visible. But if I open another thread and then return, the trace is collapsed again.

That means I have to repeatedly rediscover and reopen the fold to see content I already determined was relevant.

Example

In one example thread, the agent produced a long user-facing checklist/request-for-approval message, then followed it with a shorter terminal message saying it was blocked until the checklist was approved.

The terminal message depended on the preceding checklist to make sense. The checklist was not reasoning or tool-call output; it was user-facing response content the user needed in order to act on the terminal message.

In the collapsed view, only the short blocked message was visible. The checklist appeared only after expanding the Worked for ... trace.

Current Behavior

  1. Open the thread.
  2. Notice the assistant response appears to be missing the checklist/request-for-approval content.
  3. Expand the Worked for ... trace.
  4. The checklist and approval request become visible.
  5. Switch to another thread.
  6. Switch back.
  7. The trace is collapsed again, so the checklist is hidden again.

Desired Behavior

Add an option to preserve fold state across thread switches.

For example:

  • If I expand a Worked for ... trace in thread A, then switch to thread B, then return to thread A, that trace remains expanded.
  • If I collapse it again, that collapsed state is remembered too.
  • The preserved state could be per-thread and local to the current client/session.

Why this would help

Even if the current terminal-message folding behavior is considered correct, preserving expanded/collapsed state would make the UI predictable after a user has manually corrected the view.

It also avoids making the user repeatedly hunt for important content in long-running agent threads where the distinction between "reasoning/progress commentary" and "final/actionable assistant output" can be ambiguous.

Attachments

Screenshots available:

  • collapsed view: only the short blocked line is visible below Worked for ...
  • expanded view: the checklist and approval request are visible inside the expanded trace

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions