Desktop Code tab: auto-collapse a turn's thinking and tool calls when the task finishes (like Codex's "Worked for X min")
Status Open
Maintainer reply None cached
Activity 3 comments · opened Sep 23, 2026
Request: When a turn finishes, fold everything Claude did in that turn (thinking, tool calls and intermediate text) into one summary row, for example "Worked for 4m 12s · 23 tool calls". Only the final reply should stay expanded. Clicking the row re-expands the full trace.
Why:
- On long agent runs, the finished transcript is mostly process. Scrolling back to find earlier results is slow.
- Today the views are Normal, Thinking and Verbose, and the choice applies to the whole session. You can't follow a job live in Verbose and then have it tidy itself up when the job ends.
- Codex does this by default. It's the main readability gap when you use both tools side by side.
Suggested behaviour:
- While a turn is running, show everything live according to the current view.
- When the turn ends, collapse the process into one line showing time taken and tool-call count. The final reply stays visible.
- Add a setting to turn it on or off, for example "Collapse completed turns", next to the transcript-view default.
- Leave the current view setting unchanged.
Surface: Claude desktop app, Code tab.
🤖 Generated with Claude Code
3 Comments
+1 on that one
+1
+1. I tried building this as a mod (2.1.295, desktop, macOS):
ui.renderhooks can hide a finished turn's tool rows and interim text and add a "Worked for Ns ›" toggle above the answer. Mods can get most of the way there.What a mod can't reach is the desktop's own "Ran N commands ›" header around 2+ consecutive tool calls. For
ToolGroup, an empty tree, anisExpanded: truerewrite and a custom tree all leave it visible.<img width="237" height="93" alt="Image" src="https://github.com/user-attachments/assets/eb999efb-4fd6-4792-80dc-702fd26a96c7" />
A render hook for that header (or letting the
ToolGrouprender answer replace it) would make this fully doable as a mod.