[FEATURE] Translate user-visible thinking summaries into the configured display language

Summary

Please add an option to translate user-visible thinking summaries into the language configured for Claude Code, without requesting that Claude change its reasoning language.

This is primarily a display and localization request. A lightweight, direct translation is sufficient; polished prose is not necessary.

Problem

When I use Claude Code in Japanese, the final response can be in Japanese while the thinking summary is shown in English. Reading long English summaries during a coding session adds cognitive overhead, even when the technical content itself is simple.

For example, a summary such as:

Need to inspect the authentication middleware. Token validation may happen before routing. Check Program.cs.

could be displayed as:

認証ミドルウェアを確認する必要がある。ルーティング前にトークン検証が行われている可能性。Program.csを確認。

This level of translation would be enough. Technical terms, code, commands, and file names should remain unchanged where appropriate.

Proposed behavior

Translate only the user-visible thinking summary into the configured display language.

Conceptually:

Claude reasoning
    -> thinking summary
    -> lightweight translation for display
    -> user

Possible configuration:

{
  "language": "japanese",
  "thinkingSummaryLanguage": "auto"
}

Possible values could include:

• auto: follow the configured display language
• original: show the summary without translation
• an explicit language code such as ja

The exact configuration design is not important; automatically following the existing language setting would also solve the main problem.

Translation priorities

For thinking summaries, I would prefer:

1. Low latency
2. Preservation of technical meaning
3. Concise, direct wording
4. Natural or polished prose only as a lower priority

This should feel more like subtitles for a work log than a fully rewritten response.

Difference from changing the reasoning language

This is not a request to make Claude reason in Japanese.

The model's reasoning behavior can remain unchanged. Only the user-visible summary should be localized. Separating the reasoning language from the display language may avoid unnecessary trade-offs while making Claude Code easier to follow for non-English users.

Related issues

• #27051 appears to request control over the thinking/reasoning language itself.
• #82785 appears to concern thinking summaries not consistently following the selected language.

This proposal focuses on making the separation between reasoning and display explicit: preserve the existing reasoning behavior, then apply a lightweight translation to the user-visible summary.

If this is already intended behavior, this request can instead be treated as feedback on the desired translation style: fast, concise, and faithful rather than polished.

Why this would help

• Reduces cognitive load during long coding sessions
• Preserves technical terms and code as written
• Improves accessibility for users who prefer a non-English interface
• Does not require changing how Claude performs the underlying task