Remote Control (mobile): assistant text before/between tool calls is replaced by a summary, with no way to disable it

What happens

When driving a Claude Code session from the Claude iOS app over Remote Control, assistant text written before or between tool calls is not shown. In its place the mobile view renders a short grey summary line. Only the text written after the last tool call appears in full.

Two distinct symptoms, most likely the same root cause:

1. Text is replaced by a summary. A paragraph the assistant actually wrote is collapsed into one line.
2. The summary text is not the assistant's text. In one turn the mobile view showed a left-bar quoted block whose wording did not match anything the assistant had written — the real reply was several times longer (three numbered measurement results plus an explanation). So it is not a truncation; it is a generated summary presented where the reply should be.

Why this matters

The person using the phone cannot tell the difference between "the assistant summarised itself" and "the assistant failed to answer". In our sessions this happened repeatedly, and the reaction each time was some version of "why aren't you answering properly?" — the answer was there, the phone just didn't show it.

It also silently changes what the assistant can safely do: any turn that writes an explanation and then runs a tool loses that explanation on mobile. The workaround we ended up adopting is to put all tool calls first and the entire answer last, which is a strange constraint to design around.

Steps to reproduce

1. Start claude with Remote Control enabled.
2. Connect from the Claude iOS app.
3. Send a prompt that makes the assistant write a paragraph and then call a tool (e.g. Bash) before finishing the turn.
4. View that turn on the phone — the paragraph appears only as a summary line.

What we found looking for a setting

• The iOS app's Claude Code settings expose only text size, fonts and "wrap long lines". No option for message rendering.
• No CLI flag or settings.json key controls it either.
• From the client side, the behaviour appears to be gated purely by server-side flags (tengu_classifier_summary_kill, tengu_classifier_disabled_surfaces); there is no user-facing switch.

So a user who dislikes this has no way out, and no way to discover why their replies look truncated.

Request

1. A user-facing toggle — in the app or as a CLI/settings option — to show the assistant's full text on mobile instead of generated summaries. Even an opt-out would be enough.
2. If summaries stay the default, please make it visually clear that the line is a summary and that a longer reply exists, so it isn't mistaken for the whole answer.
3. Turkish localisation of the mobile app. The app is English-only today. The user in this case does not read English, which is a large part of why the summary line was read as a broken reply rather than as a summary — there was no wording to interpret.

Environment

• Claude Code 2.1.276, model Opus 5
• macOS host (macOS 27.0), Claude iOS app over Remote Control
• First noticed 2026-09-19; still present on 2026-09-26

This was also sent through /feedback, but that channel is one-way — filing here so it can be tracked and so others seeing the same thing can add to it.