Environment
- Platform (select one):
- Claude CLI version: 1.0.16 (Claude Code)
- Operating System: Archlinux Linux 6.14.7-arch2-1 x86_64
- Terminal: Konsole
- Python: 3.13.3
- Locale: en_US.UTF-8 (properly configured)
- Tool: Claude Code Write tool
Bug Description
The Claude Code Write tool that causes Unicode corruption, resulting in data loss for non-English content. This issue affects file operations and impacts user workflows sometimes.
Problem Summary
The Write tool occasionally corrupts Unicode characters when creating files, specifically:
- Russian Cyrillic text becomes unreadable gibberish (Dialog-First ?>4E>4** - 15+ ?0BB5@=>2 B@8335@>2 4;O 02B><0B8G5A:>3> >1=0@C65=8O 206=>9 8=D>@<0F88)
- Emoji are replaced with control characters
- English text remains unaffected
- Issue occurs "time to time" (intermittent but recurring)
Impact
This bug causes actual data corruption requiring manual file recovery. For users working with non-English content, this represents a significant reliability issue that affects productivity and trust in the tool.
Technical Details
I have prepared a comprehensive bug report with:
- Complete system environment details
- Technical analysis of the corruption pattern
- File examples (corrupted vs corrected)
- Hex dumps and encoding analysis
- Reproduction information
- Root cause hypothesis
The evidence suggests UTF-8 bytes are being incorrectly interpreted as CP1251/Windows-1251 encoding during file write operations, likely due to a race condition or intermittent encoding handling bug.
Steps to Reproduce
- Using Write() utility when writing non-Cyrillic characters (very rare)
Expected Behavior
- 🔍 Dialog-First подход - 15+ паттернов триггеров для автоматического обнаружения важной информации
- 🔄 Smart Update логика - интеллектуальная обработка изменений (ADD, UPDATE, DELETE, MERGE)
- 🗣️ Natural Language навигация - естественное взаимодействие ("покажи вчерашнее обсуждение")
Actual Behavior
- =
Dialog-First ?>4E>4 - 15+ ?0BB5@=>2 B@8335@>2 4;O 02B><0B8G5A:>3> >1=0@C65=8O 206=>9 8=D>@<0F88
- =� Smart Update ;>38:0 - 8=B5;;5:BC0;L=0O >1@01>B:0 87<5=5=89 (ADD, UPDATE, DELETE, MERGE)
- =ã� Natural Language =02830F8O - 5AB5AB25==>5 2708<>459AB285 ("?>:068 2G5@0H=55 >1AC645=85")
Additional Context
Important! Claude Code is able to fix this error on his own if he sees that something is written incorrectly in the file. But once he suspended the work saying that I was violating the rules of using the application, stating as a reason that I was allegedly trying to exploit vulnerabilities (lol)
report_unicode.md - detailed description
README.md
README_corrupted.md
Environment
Bug Description
The Claude Code Write tool that causes Unicode corruption, resulting in data loss for non-English content. This issue affects file operations and impacts user workflows sometimes.
Problem Summary
The Write tool occasionally corrupts Unicode characters when creating files, specifically:
Impact
This bug causes actual data corruption requiring manual file recovery. For users working with non-English content, this represents a significant reliability issue that affects productivity and trust in the tool.
Technical Details
I have prepared a comprehensive bug report with:
The evidence suggests UTF-8 bytes are being incorrectly interpreted as CP1251/Windows-1251 encoding during file write operations, likely due to a race condition or intermittent encoding handling bug.
Steps to Reproduce
Expected Behavior
Actual Behavior
Dialog-First ?>4E>4 - 15+ ?0BB5@=>2 B@8335@>2 4;O 02B><0B8G5A:>3> >1=0@C65=8O 206=>9 8=D>@<0F88
Additional Context
Important! Claude Code is able to fix this error on his own if he sees that something is written incorrectly in the file. But once he suspended the work saying that I was violating the rules of using the application, stating as a reason that I was allegedly trying to exploit vulnerabilities (lol)
report_unicode.md - detailed description
README.md
README_corrupted.md