Personal Preference Learning extracts preferences, styles, and behavioral patterns from current conversations and persists them for future sessions across a three-tier target architecture. It scans conversation history for explicit rules, repeated corrections, positive reinforcement, domain context, and style edits while avoiding anti-patterns like inferring from silence. The skill guides the agent through scanning, threshold evaluation, state reading, conflict detection, classification, change presentation, and approval-gated writing. It also supports a read-only query mode to inspect stored memory, view specific topics, or handle forgetting requests safely.