LLM can now flag possessive name-dropping, territorial behavior, and jealousy signals when users mention others not in the conversation. Scores feed into existing drama pipeline for warnings/mutes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>