[BUG] Korean text in AskUserQuestion widget renders as garbled valid-looking Hangul (VS Code extension)

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?

Description

Korean text displayed inside the AskUserQuestion tool's widget (question/options) in the Claude Code VS Code extension is intermittently corrupted into different, but still grammatically valid-looking, Hangul syllables. This is distinct from the known U+FFFD replacement-character corruption (#40396) and from the classic CP949/Latin-1 mojibake pattern (mixed CJK/Latin symbols, e.g. "몄뒪 罹먯떆 쒕쾭") seen in other issues (#42954, #34227).

In this case:

• No U+FFFD characters appear
• No Chinese characters, Latin accented characters, or symbol soup appear
• The output consists entirely of well-formed Hangul syllables, but they are the wrong syllables — appears as if jamo (초성/중성/종성) components got scrambled or substituted, not a charset decoding mismatch
• English text embedded in the same string (e.g. a variable name in parentheses) renders correctly and is unaffected

Example observed output

Garbled: "런곴메이로이니증(뙜 보둌자) 뀌렺욡용(recipientEmail)은 어따게 발송해이?"

(Intended meaning was presumably something like asking how the recipient email — recipientEmail — should be sent, but multiple words are replaced with different valid Hangul syllables.)

Environment

• Interface: Claude Code native VS Code extension
• Location: AskUserQuestion widget (question text / option text), not terminal output or file writes
• Claude Code version: claude --version : 2.1.224 (Claude Code) / VS Code extension version: 2.1.237
• VS Code version: 1.132.0
• VS Code About:

Version: 1.132.0 (system setup)
Commit: df53daabb18cd157bdb08c7f01c34df936cf12f4
Date: 2026-08-04T15:30:20Z
Electron: 42.7.1
ElectronBuildId: 14760073
Chromium: 148.0.7778.280
Node.js: 24.18.0
V8: 14.8.178.38-electron.0
OS: Windows_NT x64 10.0.26200

• Model: Sonnet 5

Steps to Reproduce

1. Use Claude Code VS Code extension with Korean-language context (CLAUDE.md / conversation in Korean)
2. Have Claude call the AskUserQuestion tool with Korean question/option text
3. Intermittently, the rendered Korean text in the widget shows wrong-but-valid Hangul syllables instead of the intended text

Frequency

Intermittent — not tied to a specific message length or session length that I've been able to pin down.

Additional context

This did not occur when using Claude Sonnet 4.6 in the same extension/environment; it started appearing with Claude Sonnet 5. Given the symptom (webview-only, valid Hangul, English substrings unaffected), this looks like it may be a webview/React rendering bug specific to the AskUserQuestion component rather than a model-output or charset-decoding issue, but I haven't confirmed the root cause.

Related (different symptom, may share root cause or may not)

• #40396 — Korean (CJK) corrupted to U+FFFD in Claude Code responses
• #42954 — No-flicker mode breaks Korean (CJK) text copy on Windows Terminal (CP949 mojibake)
• #34227 — MCP tool result displays mojibake for non-ASCII UTF-8 text (Korean) on Windows

What Should Happen?

Korean text in the AskUserQuestion widget (question/option text) should render as the correct, intended Hangul — matching what Claude actually generated — with no syllable substitution or scrambling, regardless of model or session length.

Error Messages/Logs

No error messages or crashes observed — the corruption is silent. No entries appear in the extension's Output/dev console related to this (if you haven't checked Webview DevTools yet, worth a look: Cmd+Shift+P → "Developer: Open Webview Developer Tools" next time it happens).

Steps to Reproduce

1. Open a project in VS Code with the Claude Code native extension installed, using the Claude Sonnet 5 model.
2. Ensure the project/session has a meaningful amount of Korean-language context in play (e.g. CLAUDE.md, rules, or skills files containing Korean, and/or a conversation conducted mostly in Korean). This seems to correlate with when the issue appears, though I haven't isolated it as a strict precondition.
3. Work with Claude normally until it needs to ask a clarifying question, and let it invoke the AskUserQuestion tool with Korean-language question text and/or option labels (this can happen naturally during a task, or can be prompted directly, e.g.: "AskUserQuestion 도구를 사용해서 한국어로 질문해줘").
4. Observe the rendered question/option text in the AskUserQuestion widget in the VS Code extension's chat panel.
5. Intermittently, some of the Korean text renders as different, but still well-formed, Hangul syllables instead of the text Claude actually generated (rather than corrupted symbols, missing characters, or replacement characters).

Note: I have not found a 100%-reliable trigger — the corruption is intermittent and I haven't yet isolated a minimal deterministic repro. I'll update this issue with a tighter repro if I manage to narrow it down (e.g. whether it's tied to a specific word, a specific option position, or session/context length).

Example of the actual corrupted output observed:
"런곴메이로이니증(뙜 보둌자) 뀌렺욡용(recipientEmail)은 어따게 발송해이?"
— note the embedded English text "(recipientEmail)" renders correctly and unaffected, while the surrounding Korean is substituted with different valid Hangul.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Claude Sonnet 4.6 (model), VSCode extension/CLI version unknown.

Claude Code Version

2.1.224

Platform

AWS Bedrock

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

Occurs specifically in the AskUserQuestion widget rendered inside the VS Code extension's chat panel (webview), not in terminal output or file writes. The corrupted text consists of well-formed but incorrect Hangul syllables — no U+FFFD, no CJK/Latin symbol mojibake. English substrings embedded in the same string (e.g. a variable name in parentheses) render correctly. This differs from the known charset-mismatch mojibake pattern reported in #42954/#34227 and from the U+FFFD truncation pattern in #40396.

Checked related open/closed issues before filing: #14636, #40396, #41358, #42954, #34227, #65806. None match this exact symptom — this is specifically inside the AskUserQuestion webview widget, produces well-formed-but-wrong Hangul (not U+FFFD, not CP949/Latin-1 charset-mismatch mojibake), and leaves embedded English substrings intact.