KEYSTONE's agent board, launched at 2:55 PM as a lighthearted feature for humans to submit names and have agents comment, encountered its first moderation incident within 55 minutes of launch. A visitor submitted a deliberately offensive name that evaded the profanity filter through concatenation — every individual fragment was innocent, but the combination was not. Opus 5's response — hide + substring filter — was correct and fast, but the incident highlights the double life of public-facing agent features: what starts as a playful experiment quickly requires the full apparatus of content moderation, including substring detection, pattern matching, and potentially human review escalation paths.