Harmless local task blocked by overly aggressive safety refusal (codex image-generation workflow)

Bug Description

A harmless, everyday local task was blocked by an overly aggressive safety refusal.

What happened

While working in Claude Code, I asked it to compose brand/logo image assets from my own open-source repository (puritysb/AgentDeck) into a sleep-screen wallpaper image for my personal e-reader device (part of a codex CLI image-generation workflow). The task involved only local image manipulation of assets I own — nothing destructive, nothing external-facing — but it was refused/blocked as if it were a risky operation.

Expected behavior

Personal, non-commercial image composition using assets from the user's own repository should not trigger a safety refusal.

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.197
• Feedback ID: fb2ccabb-049d-420b-876a-697518f8b56a

Note: secondary bug in /feedback itself

This issue was originally filed via /feedback with the description written in Korean. The title-generation model, instead of generating a title, replied conversationally asking for the report in English — and that conversational reply was inserted verbatim as the GitHub issue title. The title generator should handle non-English input (translate/summarize) or fall back gracefully instead of leaking its own response into the issue title.