Desktop app (Windows): Transcript view "Thinking" never renders thinking — blocks exist in .jsonl; showThinkingSummaries:true has no effect

Summary

In the Claude Code desktop app on Windows, setting the Transcript view to Thinking never displays any thinking/reasoning content — not on old messages and not on brand-new streamed replies — even though the session transcripts on disk provably contain thinking blocks, and "showThinkingSummaries": true is set in ~/.claude/settings.json (app fully restarted afterwards).

Environment

• Claude Code desktop app, Windows 11 Pro (10.0.26100)
• CLI version: 2.1.150
• Models tried: Opus 4.8 (effort high; also with fast mode) and Sonnet
• Settings: "showThinkingSummaries": true in ~/.claude/settings.json (verified valid JSON, app restarted)

Evidence that thinking IS produced

Counted "type":"thinking" blocks in the session .jsonl transcripts (~/.claude/projects/<cwd>/):

• Session A: 124 thinking lines / 1,147 rows (includes the latest assistant turns)
• Session B: 4,852 thinking lines / 35,473 rows
• Session C: 1,811 thinking lines / 12,706 rows
• Sessions started AFTER enabling showThinkingSummaries + full app restart also contain thinking blocks — still nothing rendered.

Steps to reproduce

1. Desktop app on Windows, any session.
2. Set Transcript view to "Thinking".
3. Send a message; watch the new reply stream in.
4. No thinking section appears (no collapsible block, nothing). Transcript .jsonl for the same turn contains a "type":"thinking" block.
5. Add "showThinkingSummaries": true to ~/.claude/settings.json, restart app, new session → same result.

Expected

Thinking view shows the reasoning (or a summary of it) for turns that have thinking blocks.

Actual

Thinking view is indistinguishable from Normal — no thinking ever rendered.