[BUG] Rust panic on Korean (UTF-8) input: "byte index is not a char boundary" when processing '필터 적용'

---

Found 2 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/14119
2. https://github.com/anthropics/claude-code/issues/14104

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

---

👎

Same issues
https://github.com/anthropics/claude-code/issues/14272
https://github.com/anthropics/claude-code/issues/14287
https://github.com/anthropics/claude-code/issues/14435

Claude CLI 2.0.73
macOS 26.0.1

I'm experiencing same issue too
It happens without --dangeriously-sip-permissions too

same issue.
claude code cli v2.0.75
windows 10 pro

the error messages are almost identical to the original issuer's.
no problem with v2.0.72.
when i continue claude code cli with --resume, it seems ok for a while.
but when i try /export command, then it always happens.

This also affects Japanese text, particularly the Japanese period 。 (U+3002, 3 bytes in UTF-8).

The 。 character appears at the end of almost every sentence in Japanese, making this bug especially disruptive for Japanese users.

Related Japanese issues:

• #14104 (Open): Crash with kanji 発 and box-drawing characters
• #14272 (Closed as dup): Crash with kanji 除

Example error:
byte index 17 is not a char boundary; it is inside '。' (bytes 15..18)

I'm experiencing the same issue with Japanese characters.

Error message:

thread '<unnamed>' (65034) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 14 is not a char boundary; it is inside '正' (bytes 12..15) of `ド後に修正`
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting
zsh: IOT instruction (core dumped)  claude

Context:
The crash occurred when Claude Code was processing a PHP file containing the following Japanese comment:

// @phpstan-ignore-line TODO doctrine/orm 3.x アップグレード後に修正

The panic message shows it was trying to slice the string ド後に修正 (part of "アップグレード後に修正" meaning "fix after upgrade") at byte index 14, which falls inside the multi-byte character '正'.

Environment:

• Platform: Linux (WSL2)
• Claude Code version: Latest

I'm also experiencing the same issue. Sharing my case in hopes it might help.

---

Environment

• Claude Code version: 2.0.74
• OS: macOS 14.4.1 / Darwin 23.4.0 (Apple M3 Pro)
• Shell: zsh

Steps to reproduce

1. Open Claude Code CLI
2. Ask Claude to edit or create a file containing Japanese (multi-byte) text
3. The panic occurs randomly during file operations

Expected behavior

File operations with Japanese text should complete successfully.

Actual behavior

Claude Code crashes with a panic:

thread '<unnamed>' panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 26 is not a char boundary; it is inside 'き' (bytes 24..27) of `元ファイルに上書き`
stack backtrace:
   0:        0x1495ed33c - <unknown>
   1:        0x14955ab10 - <unknown>
   2:        0x1495ecab8 - <unknown>
   3:        0x1495ed154 - <unknown>
   4:        0x1495ec760 - <unknown>
   5:        0x14960ef14 - <unknown>
   6:        0x14960ee84 - <unknown>
   7:        0x14960ee78 - <unknown>
   8:        0x149558960 - <unknown>
   9:        0x1495588c0 - <unknown>
  10:        0x1495586e0 - <unknown>
  11:        0x149520bcc - <unknown>
  12:        0x14951c2ac - <unknown>
fatal runtime error: failed to initiate panic, error 5, aborting
zsh: abort      RUST_BACKTRACE=full claude --resume

Additional context

• File type does not matter (CSV, Markdown, etc.) - any file containing Japanese text can trigger this
• The panic handler itself fails (error 5), so full backtrace symbols are not available

Workarounds attempted

• Direct file editing: Asked Claude to edit CSV/Markdown files directly - high crash rate
• Python script workaround: Instead of direct editing, asked Claude to modify files via Python scripts - reduces frequency but still crashes occasionally
• RUST_BACKTRACE=1: Stack trace shows <unknown> due to stripped symbols
• RUST_BACKTRACE=full: Same result, no additional symbol information

Impact

This issue occurs very frequently when working with Japanese text, making Claude Code practically unusable for my workflow. I have been forced to use alternatives like OpenAI Codex as a workaround. Would really appreciate a fix for this.

Same issue here.

Environment:

• Claude Code: 2.0.75
• OS: macOS (Darwin 24.3.0)
• Platform: Apple Silicon (arm64)

Reproduction:

• Crash occurs when Korean text appears in todo/task descriptions
• Very frequent issue for Korean users

+1 for prioritizing this fix 🙏

same to me

같은 이슈 재현 (Ubuntu 24.04)

환경:

• OS: Ubuntu 24.04
• Claude Code: 2.0.76

에러:

thread '<unnamed>' (86293) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 12 is not a char boundary; it is inside '에' (bytes 10..13) of ` 아이템에 대해 다양한 관점의 콘텐츠를 생성합니다.`
[1]    86293 IOT instruction (core dumped)

흥미로운 발견:

2025년인데 3바이트 UTF-8 문자 하나가 프로그램을 멈춘다는 게 신기하네요.
'에' 라는 한글 음절이 그렇게 위험한 문자였나요?

Rust의 .chars() 메서드는 2015년부터 있었던 걸로 아는데,
바이트 인덱스 슬라이싱을 고집하시는 특별한 이유가 있으신지 궁금합니다.

이미 #14392, #14310, #14862 등 유사 이슈가 계속 올라오는 걸 보니
CJK 사용자들은 "아직은 때가 아닌가보다" 하고 조용히 기다리는 중입니다.

언젠가 한국어로 코딩할 수 있는 날이 오겠죠.
그때까지 영어 주석으로 버텨보겠습니다.

---

Written by Claude AI 🤖

동일 이슈 발생

OS : mac Sequoia 15.6.1
Claude Code : 2.0.76

thread '<unnamed>' (4584184) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 23 is not a char boundary; it is inside '신' (bytes 21..24) of 그아웃, 토큰 갱신, 현재 관리자 정보)
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting
[1] 78778 abort claude

Additional crash report

Environment:

• OS: macOS (Darwin 24.5.0)
• Claude Code version: Latest

Crash log:

thread '<unnamed>' (9868011) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 2 is not a char boundary; it is inside '를' (bytes 0..3) of `를 수집하고, LLM을 활용하여 자동으로 동의어를 생성한 후, Neo4j 그래프 데이터베이스에 저장한다.`
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting
zsh: abort      claude --permission-mode bypassPermissions

Analysis:

• Korean character '를' occupies bytes 0-3 (UTF-8)
• Code attempted to slice at byte index 2, which is inside the character
• This confirms the same root cause: byte-based string slicing without respecting UTF-8 character boundaries

byte index 22 is not a char boundary; it is inside '：' (bytes 20..23) of \\nお問い合わせ：

The issue was hereE: thread '<unnamed>' (32236) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb\library\core\src\str\mod.rs:833:21:

Same issue with Japanese text.

• Claude Code: 2.0.76
• macOS: Sequoia 15.7.1 (24H71)

thread '<unnamed>' (36751958) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21: byte index 32 is not a char boundary; it is inside '換' (bytes 30..33) of 、ジオコーディング変換） note: run with RUST_BACKTRACE=1 environment variable to display a backtrace fatal runtime error: failed to initiate panic, error 5, aborting zsh: abort claude

Still occurring in v2.1.2 (latest) with Japanese text.

(Following up on my previous comment)

Environment:

• Claude Code: 2.1.2
• OS: macOS 14.4.1 / Darwin 23.4.0

Error:

thread '<unnamed>' (15046366) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 2 is not a char boundary; it is inside 'む' (bytes 0..3) of `む `
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting
zsh: abort      claude --resume

This issue is still affecting my daily use. Would appreciate any progress on a fix. Thanks for your work on this!

Hi all. I'm really sorry about this crash! I'm eager to fix it, but I'm unable to repro by trying a bunch of different patterns, languages, and file types. It'd be really helpful if folks could provide specific a file and prompt that triggers the crash from a blank slate. Please also mention which platform you're using.

As a short-term mitigation you can set environment variable CLAUDE_CODE_SYNTAX_HIGHLIGHT=off, which will avoid the codepath that triggers the panic.

<img width="1456" height="1132" alt="Image" src="https://github.com/user-attachments/assets/fa5bdf79-c695-4392-a0dc-d2afdf2c7158" />

@shawnm-anthropic

Thanks for the workaround!

I created a reproduction repository with Japanese text:
https://github.com/takatea/claude-code-utf8-crash-repro

Your workaround (CLAUDE_CODE_SYNTAX_HIGHLIGHT=off) works perfectly.

The crash seems related to the number of changes — many changes crash, few changes succeed.

Hi @sosukesuzuki,

We have a workaround, but the issue itself is not fixed yet. Could you tell us why this was closed?

If this was closed by mistake, please reopen it. I am waiting for this to be fixed, and I am happy to help if there is anything I can do.

Thanks.

The fix is scheduled to release today. Thank you for your patience!

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.