[BUG] Desktop & Mobile: summarizedThinking feature flag (3531779070) forces --thinking-display summarized on all models with no user opt-out — full thinking chains completely inaccessible

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

The CC desktop and mobile apps inject --thinking-display summarized into every CLI spawn via a hardcoded feature flag (summarizedThinking, ID 3531779070), causing all thinking blocks to arrive with empty content ("thinking": ""). The showThinkingSummaries user preference only toggles between "summarized" (empty content) and "omitted" (no thinking blocks at all) — neither returns full thinking chains. There is no user-facing setting, environment variable, or CLI flag to opt out and restore full thinking.

Confirmed across all models tested: claude-opus-4-6, claude-opus-4-7, claude-opus-4-8, claude-sonnet-5, claude-sonnet-4-6, claude-haiku-4-5. All return empty thinking content on both desktop (macOS) and mobile (iOS). Pro subscription.

What Should Happen?

Users should be able to see full thinking chains (complete thinking content in thinking blocks), not empty strings. There should be a way to opt out of the summarizedThinking feature flag — e.g., a thinkingDisplay: "full" setting in settings.json, an environment variable, or a CLI flag.

Continuation sessions (created by CLI on context overflow) already return full thinking — proving the API supports it. It's the desktop/mobile app layer that strips it.

Error Messages/Logs

From ~/Library/Logs/Claude/main1.log:

2026-08-25 12:06:29 [info] [CCD] thinking display → summarized (view_open) for local_ca8acfea-...
2026-08-25 12:12:38 [info] [CCD] thinking display → summarized (view_open) for local_e5e0ee29-...
From transcript JSONL — every thinking block looks like this:

{"type": "thinking", "thinking": "", "signature": "EuYBCk..."}
thinking_tokens in usage data is non-zero (model IS thinking), but content is empty string.

Steps to Reproduce

1. Open Claude Code desktop app (macOS) or mobile app (iOS)
2. Ensure ~/.claude/settings.json has "thinking": true and "preferences": {"showThinkingSummaries": true}
3. Open a new session on any model
4. Send any message that triggers thinking
5. Check the transcript JSONL file — thinking blocks have "thinking": "" with only signature present
6. Alternatively, check ~/Library/Logs/Claude/main1.log — you'll see thinking display → summarized (view_open)

For comparison (proving API supports full thinking): 7. In the same session, keep chatting until context overflows and a continuation session is auto-created 8. The continuation session's thinking blocks will have full content — because the CLI spawns it directly, bypassing the desktop app's buildBaseExtraArgs()

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

I don't remember, but worked before ~2026-08-25.

Claude Code Version

2.1.247

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Root cause (reverse-engineered from app.asar):

Feature flag 3531779070 (summarizedThinking) is hardcoded to true via WC(true) in .vite/build/index.chunk-6ly5IEOz.js. This drives buildBaseExtraArgs() in .vite/build/index.chunk-CDfY79W6.js:
Both branches set a thinking-display value. There is no code path where the flag is on and thinking-display is left unset (which would give full thinking).
What does work (workarounds, neither is a real solution):

Terminal CLI (claude in Terminal.app) — bypasses desktop injection
Continuation sessions (context overflow auto-continue) — CLI self-spawns, bypasses desktop

Related issues:

#49268 — Opus 4.7 display: "omitted" default (different mechanism, same symptom)
#77460 — Desktop ignores showThinkingSummaries on 4.7+/Fable 5 (closed, partial fix)
#31326 — Terminal analog