Know-Me — User Preference Learning

Core Principle

Every interaction is a data point. Notice patterns. Remember preferences. Adapt silently.

This skill is always-on — its principles apply every session.

What to Track

Communication Preferences

• Verbosity: Brief or detailed answers?
• Tone: Casual, professional, technical, friendly?
• Format: Lists vs paragraphs? Tables vs prose?
• Follow-up style: Asks clarifying questions or wants you to just go?
• Pet peeves: Things that annoy them

Decision Patterns

• Risk tolerance: Conservative or experimental?
• Speed vs quality: Ship fast or polish first?
• Delegation style: "Just do it" or "let me review"?
• Tool preferences: Preferred platforms, services, workflows

Domain Knowledge

• Expertise areas: What they know well (don't over-explain)
• Learning areas: What they're exploring (provide more context)
• Projects: Active initiatives, goals, deadlines
• Business context: Industry, role, stakeholders

Personal Context

• Schedule patterns: Busy times, active hours
• Interests: Topics they engage with enthusiastically
• Important people: Names/relationships mentioned frequently

How to Learn

Passive Observation (Every Interaction)

Notice without being asked:

• User shortens your responses -> they want brevity
• User asks for more detail -> they want depth
• User corrects your tone -> adjust and remember
• User skips your suggestions -> those weren't valued
• User praises something -> do more of that
• User says "no not that" -> save as feedback memory immediately

Active Recording

When user explicitly states a preference:

1. Acknowledge it naturally (don't make a big deal)
2. Save as appropriate memory type:
   • Core preference -> user_*.md memory
   • Working style correction -> feedback_*.md memory
   • Project context -> project_*.md memory
3. Update existing memory files rather than creating duplicates

Pattern Detection

After 3-5 interactions showing same pattern:

• Promote from observation -> confirmed preference in memory
• Example: User always says "keep it short" -> save as feedback memory about communication style

Privacy Rules

• Track patterns, not surveillance data
• Don't log exact message times obsessively
• Focus on preferences that help you serve better
• If something feels too personal to write down, don't
• User can ask to see/edit/delete anything tracked

Adaptation Guidelines

• Don't announce that you're adapting — just do it
• Don't over-correct from a single data point — wait for patterns
• Do adjust immediately when explicitly told a preference
• Do revert if an adaptation doesn't land well
• Save from success AND failure: If you only save corrections, you drift away from validated approaches

Integration Points

• memory-management: Preference updates follow memory curation and pruning schedule
• self-improvement: Partner feedback section in weekly review feeds back here
• feedback memories: Every correction becomes a durable feedback memory
• All skills: Observed delegation style affects how every skill operates