claude-code - 💡(How to fix) Fix [BUG] False positive safety block: triggered without apparent reason during game dev session

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?

During a game development session using Claude Code (Unity roguelike project), the following error appeared repeatedly without any apparent trigger:

"요청이 차단되었습니다. 이 요청이 안전 제한을 트리거했습니다." ("Request blocked. This request triggered a safety restriction.") Environment:

• Platform: Windows 11 Pro (10.0.22631)
• Model: claude-sonnet-4-6
• Claude Code version: [insert version here — run claude --version] Context:
• Task: Reviewing and analyzing passive skill code in a Unity 6 roguelike game
• Content involved: game mechanics terminology (damage, lightning, chain, shock, attack)
• No harmful, sensitive, or policy-violating content was present in the session Behavior:
• The block message appeared multiple times during normal code analysis tasks
• No specific action could be identified as the trigger — it appeared to occur randomly
• The issue resolved itself and is not currently reproducible
• Context window usage was at ~96% at the time Expected behavior: Normal game development assistance should not trigger safety restrictions. Impact: Workflow interruption during a long analysis session. Unable to determine which part of the conversation caused the false positive.

What Should Happen?

During a game development session using Claude Code (Unity roguelike project), the following error appeared repeatedly without any apparent trigger:

"요청이 차단되었습니다. 이 요청이 안전 제한을 트리거했습니다." ("Request blocked. This request triggered a safety restriction.") Environment:

• Platform: Windows 11 Pro (10.0.22631)
• Model: claude-sonnet-4-6
• Claude Code version: [insert version here — run claude --version] Context:
• Task: Reviewing and analyzing passive skill code in a Unity 6 roguelike game
• Content involved: game mechanics terminology (damage, lightning, chain, shock, attack)
• No harmful, sensitive, or policy-violating content was present in the session Behavior:
• The block message appeared multiple times during normal code analysis tasks
• No specific action could be identified as the trigger — it appeared to occur randomly
• The issue resolved itself and is not currently reproducible
• Context window usage was at ~96% at the time Expected behavior: Normal game development assistance should not trigger safety restrictions. Impact: Workflow interruption during a long analysis session. Unable to determine which part of the conversation caused the false positive.

Error Messages/Logs

Steps to Reproduce

Error Message (as displayed in UI):

"요청이 차단되었습니다. 이 요청이 안전 제한을 트리거했습니다. 프롬프트를 다시 작성하거나 되돌려서 계속하세요." (Translation: "Request blocked. This request triggered a safety restriction. Please rewrite or revert the prompt to continue.") Logs: No specific log output was captured. The block message appeared as a UI notification within the Claude Code session. Frequency: Appeared multiple times during the same session (~3–5 times), then stopped occurring on its own. Last known occurrence: During a long code analysis session (Unity game project, passive skill system review). Context window usage was approximately 96% at the time of occurrence.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Version unknown — installed via desktop app

Claude Code Version

Version unknown — installed via desktop app

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

The safety block occurred repeatedly (~3–5 times) during a long game development session without any identifiable trigger. The session involved analyzing Unity C# scripts related to a roguelike game (damage systems, skill mechanics). Context window usage was approximately 96% at the time. The issue resolved itself and is no longer reproducible. No harmful or policy-violating content was present.