[BUG] Desktop Code tab: thinking only shows as a one-line summary, and some sessions launch with --thinking-display omitted and show none

Summary

In the Desktop app's Code tab, I can't see the model's thinking in any session. This isn't limited to forks.

1. Normal sessions: thinking shows only as a single collapsed one-line summary per turn (e.g. "This person wants to report a genuine bug about missing thinking blocks in the UI, which is legitimately h… ›"). There's no way to see more than that line.
2. Some sessions show no thinking at all. Their CLI process is launched with --thinking-display omitted. This happened to a user fork and to a session started by another session. A second user fork made later got summarized, so the trigger isn't simply "fork".

Environment

• Claude desktop app 2.42304.0 (Code tab)
• macOS 27.0.1 (build 26A434)
• Model: Opus 5.5 (claude-opus-5-5), effort: medium
• Permission mode: auto; Remote Control: on
• Transcript view: Normal and Thinking both tried

Steps to reproduce

1. Open the Code tab, start a session, and send a prompt that needs reasoning.
2. Look at the thinking row above the reply: it's a single summary line.
3. Fork the session, or let a session start another session, and send a prompt there. Sometimes no thinking row appears at all.

Expected: thinking is viewable in every session, with a way to see more than one summary line.
Actual: one summary line at most, and in some sessions nothing.

Evidence (process metadata only, captured 2026-10-15)

Running claude processes on this machine and their --thinking-display value:

| Session | How it started | --thinking-display |
|---|---|---|
| Main session (local_db8ab037…) and other normal sessions | New session | summarized |
| local_adddb5d3… | User fork of the main session | omitted |
| local_27b44bca… | Started by another session | omitted (main.log later shows thinking display → summarized (view_open), but the process still has omitted) |
| local_56c4d0ae… | Later user fork of the main session | summarized |

Side issue: false-positive safety flags while debugging

While asking Claude to help diagnose this (checking process flags and app logs), several messages were blocked with "Message flagged … Details: [reasoning_extraction]". I was only trying to troubleshoot this display bug. A supported way to check whether thinking is being requested and displayed would make bugs like this one easier to report.

Possibly related

#92535 (one-line summaries only), #94032 (summarized thinking forced with no opt-out), #79021 (thinking present but not rendered)