Action-summary preview text generated in Japanese instead of Korean
Description
While using Claude Code in the mobile app (chat/action-summary preview panel), the auto-generated "what I'm doing" summary text that appears above tool-call results was rendered in Japanese instead of matching the conversation language (Korean). The user was chatting entirely in Korean, and my (the model's) actual text replies were in Korean, but the short preview/summary blurb shown while tool calls were running came out in Japanese (e.g. "良い質問です、実際に確認してみます...").
This is not something the model wrote as a real reply — it looks like a separate, smaller summarization step (used to render a live "doing X..." preview during tool execution) picked the wrong output language.
Steps to reproduce
- Use Claude Code from the mobile app, in a Korean-language conversation.
- Ask a question that triggers multiple tool calls in a row (in this case: several Bash/Read calls while investigating a bug).
- Observe the live action-summary preview text shown above the tool-call results while they run.
Expected behavior
The summary/preview text should be in the same language as the conversation (Korean), consistent with the rest of the UI and the model's actual replies.
Actual behavior
The summary/preview text was written in Japanese, while everything else (my own prior/following replies, UI labels) was in Korean.
Environment
- Client: Claude Code mobile app (Android)
- Model: Sonnet 5 (claude-sonnet-5)
- Conversation language: Korean
Additional request
This is a client/UI-level bug unrelated to my own project code, so it cost me API/output tokens without any way for me to have prevented or fixed it. I'd like to request a token usage credit/refund for the turns affected by this bug. I understand issue tracker isn't the billing channel, but wanted to flag the request here as well per Claude's suggestion — happy to follow up through the proper support channel too.