dvcrn 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
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:
🎨 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:
🎨 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) |