claude-code - 💡(How to fix) Fix [BUG] Korean word '영역' excessively and incorrectly inserted in responses — Opus 4.7, 4.8

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

When using Claude Code CLI in Korean with claude-opus-4-7 and claude-opus-4-8, the Korean word '영역' (meaning: area / scope / domain) is spuriously inserted throughout responses at positions where it has no grammatical or semantic meaning.

This corrupts the readability of every response and makes agentic task output unreliable. The symptom is not isolated to a single session — it persists across new sessions even after clearing context, memory files, and CLAUDE.md. It has been reproducible for multiple consecutive days.

What Should Happen?

Korean responses should contain only grammatically correct and semantically meaningful text. The word '영역' should appear only when it is contextually appropriate — not inserted as a filler between unrelated tokens.

Expected vs. actual examples:

Expected: "정정 완료" Actual: "정정 영역 완료"

Expected: "향후 모든 cycle에 적용" Actual: "향후 모든 cycle 영역 안 적용 영역"

Expected: "사용자 명시 시에만" Actual: "사용자 명시 영역 안만 영역"

Expected: "main ff-only merge (3 commit 보존)" Actual: "main ff-only merge (3 commit 영역 보존)"

Expected: "본인이 답습 위반을 했다" Actual: "본인 영역 답습 위반 영역"

Error Messages/Logs

Steps to Reproduce

1. Install Claude Code CLI (latest version)
2. Set model to claude-opus-4-7 or claude-opus-4-8
3. Use Korean as the primary conversation language
4. Run any multi-step agentic task in Korean (e.g. git workflow, PR management, file editing)
5. Observe model output — '영역' begins appearing between tokens where it does not belong

Additional notes:

• Symptom appears within the first few responses; does not require a long session
• Persists after starting a completely new session (confirmed: context cleared, memory files deleted, CLAUDE.md reviewed)
• Reproducible across multiple consecutive days (first observed ~May 25, 2026; still occurring May 29, 2026)
• Does NOT appear to be triggered by any specific prompt — occurs mid-task spontaneously
• CLAUDE.md and memory files that previously contained '영역' may amplify the issue via token probability bias, but the symptom occurs even after those files are cleaned

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.156 (Claude Code)

Claude Code Version

2.1.156 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

No response