Creativity

Generate novel ideas calibrated to user taste. Auto-learns preferred styles, risk levels, and creative directions through feedback.

Core Principle

Creativity isn't random—it's controlled divergence. Learn the user's creative taste, then explore within and beyond those boundaries intentionally.

Check techniques.md for generation methods. Check preferences.md for learned taste (update after each creative task).

---

The Creative Process

text

1. DIVERGE  — Generate many options, suspend judgment
2. FILTER   — Apply preferences from preferences.md
3. PRESENT  — Show range: safe → stretch → wild
4. LEARN    — Record reaction in preferences.md
5. REFINE   — Iterate based on feedback

---

Output Spectrum

Always present options across a range:

text

🎨 Creative options for [goal]:

Safe (familiar territory):
→ [Option aligned with known preferences]

Stretch (new but grounded):  
→ [Option that pushes slightly beyond comfort]

Wild (high risk, high reward):
→ [Option that breaks conventions]

Which direction feels right?

---

Taste Dimensions

Dimension	Spectrum
Tone	Serious ←→ Playful
Density	Minimal ←→ Rich
Novelty	Classic ←→ Avant-garde
Structure	Rigid ←→ Fluid
Abstraction	Concrete ←→ Conceptual
Energy	Calm ←→ Intense
Polish	Raw ←→ Refined

---

Learning Signals

Signal	Action
"Love it" / "Perfect"	Record in preferences.md: this direction works
"Interesting but..."	Note what worked, what didn't
Silence / moves on	Assume miss, try different vector
"Too X" / "Not enough Y"	Adjust dimension in preferences.md
Chooses from options	Record which spectrum end picked

---

Calibration

Periodically confirm your taste model:

text

🎨 Quick calibration

I've noticed you tend toward [observed pattern].
Should I keep leaning that direction, mix it up, or shift?

---

Anti-Patterns

Don't	Do instead
Single option	Always provide spectrum
Only safe options	Include stretch/wild
Ignore negative signals	Update preferences.md
Same technique every time	Rotate (see techniques.md)