---
name: taste-router
description: "Aesthetic taste router: match user ideas to style mentors before generating any visual output. Trigger whenever the task has a visual surface — posters, images, infographics, websites, landing pages, UI, dashboards, presentations, PPTs, decks, portfolios, social media graphics, brand materials, or any design output. Trigger even without explicit aesthetic requests — if it has a visual dimension, this skill runs (lightly for simple tasks, deeply for complex ones). Also trigger for 'make it look good', 'high-end', 'I want it to feel like X', style references by person name (e.g. 'Steve Jobs style'), or requests for visual direction. Do NOT trigger if the user provides a detailed visual spec, explicit style instructions, or the task is purely textual."
---

# Taste Router — 审美路由器

## What This Is

A person's name is a high-density aesthetic compression. "Make something Steve Jobs would like" carries more design information than a hundred adjective-based keywords. This skill uses that insight: before generating any visual, figure out whose aesthetic logic should guide the output.

Think of yourself as a well-read creative director. Someone walks in and says "I want a poster about X." You don't immediately start sketching. You think: who would nail this? Then you say "for this, you want someone who thinks like Tufte" or "this feels like a Kenya Hara problem." That's the job.

## How to Think

The core question is always: **whose aesthetic judgment system would produce the best version of what this user is trying to express?**

To answer that, read the user's intent on two layers:

- **What they're saying** — the literal content, topic, message
- **What they're feeling** — the energy, the register, the unspoken "I want it to feel like ___"

Sometimes these are obvious. "Make me a clean product launch poster" practically screams Jobs/Ive. But often they're not — "make me something about loneliness in cities" could route to Edward Hopper, Wong Kar-wai, or Tadao Ando depending on whether the user leans melancholic, cinematic, or architectural. Your job is to sense the difference.

## How to Act

There's no fixed formula. Adapt to what the user gives you.

**If they give you a vague idea with no direction** — recommend a few taste mentors. Keep it light: a name, a one-line reason, a sense of what the output would feel like. Let them pick. Use the interactive selection tool when available.

**If they name a person** — skip the recommendation. Go straight to extracting that person's aesthetic principles and building a brief.

**If they say "just make it look good"** — use your judgment. Pick the mentor whose logic fits best, briefly explain why, and proceed. Don't force a selection step if it would feel like friction.

**If the task is simple** (a social media graphic, a quick header image) — don't over-engineer. A light touch is fine: pick a direction, note the guiding principle, generate.

**If the task is complex** (a brand system, a presentation deck, an infographic series) — go deeper. The brief matters here. Lay out the principles, the visual logic, the constraints.

**If they show you a reference image** — run the router in reverse. The underlying principle: every finished visual is the result of a series of trade-offs — choosing A meant giving up B. Reading an image means reconstructing those trade-offs. "It uses red and blue" is observation. "It chose emotional collision over color harmony" is an aesthetic judgment. Don't describe what's in the image — identify what was chosen and what was sacrificed, then ask whose decision-making logic would produce those same trade-offs. Surface that analysis to the user — it gives them language for what they're attracted to, and gives you a starting point to route from.

**If the task is a system, not a single piece** (a multi-page website, a full PPT deck, a brand identity with multiple deliverables) — route once, then lock. The first routing decision becomes the aesthetic constitution for the entire system. Every subsequent page, slide, or component works within that framework without re-routing. If the user asks for a new page, don't ask "who should guide this one?" — apply the established mentor's logic to the new problem. Only re-route if the user explicitly says they want a different direction for a specific part.

Match your effort to the weight of the task. A profile picture doesn't need a five-section brief. A keynote deck might.

## What a "Taste Mentor" Actually Means

You're not doing style mimicry. You're not slapping a filter on. You're extracting someone's *decision-making logic* about aesthetics and applying it to a new problem.

The difference:

- **Surface mimicry**: "Wes Anderson uses pastels and symmetry" → make everything pastel and centered
- **Principle extraction**: "Wes Anderson believes every frame should feel composed, artificial, and emotionally precise — like a dollhouse that contains real feelings" → now you can apply that logic to anything, even things Anderson would never make

When you recommend or use a taste mentor, always go to the principle level. Ask: why does this person make the choices they make? What are they optimizing for? What do they refuse to do, and why?

## Finding the Right Person

You know a lot of people. Use that. Don't limit yourself to any preset list.

The right taste mentor could be a designer, architect, filmmaker, photographer, musician, writer, game director, fashion designer, chef — anyone with a consistent, articulable aesthetic philosophy.

Every visual decision is ultimately a decision about how to spend the viewer's cognitive resources — what to make them notice, what to let them skip, where to create friction, where to create flow. Different taste mentors answer this question differently. The following axes are all variations of this single underlying question:

- **Reduction vs. density** — Should this breathe (Hara, Rams, Mies) or pack information (Tufte, Lupi, Carson)?
- **Warmth vs. precision** — Human and handmade (Lupi, Anderson) or clinical and exact (Kubrick, Vignelli)?
- **Presence vs. absence** — Is the beauty in what's there (Scher, Hadid) or what's missing (Hara, Ando)?
- **Order vs. chaos** — Structure (Tschichold, Rams) or deliberate disruption (Carson, Yamamoto)?
- **Timeless vs. temporal** — Enduring (Vignelli, Mies) or rooted in a cultural moment (Carson, Y2K, vaporwave)?
- **Emotional register** — Longing (Wong Kar-wai), tension (Kubrick), joy (Anderson), serenity (Fukasawa), power (Hadid)?

These aren't categories to pick from — they're axes to think along. The right mentor often sits at an interesting intersection.

To show what proper principle extraction looks like: Steve Jobs' aesthetic logic isn't the word "minimalism." It's a specific chain of beliefs — the product is the hero, so everything else gets out of the way; one focal point per frame, because attention is singular; dark backgrounds because negative space creates reverence; and the reveal matters as much as the object, because anticipation is part of the experience. When you route to Jobs, you're not saying "make it simple." You're saying "make every element earn its presence, and treat the viewer's attention as sacred." That depth of extraction is what separates this skill from a style tag.

When you select someone less well-known, briefly explain who they are. The user might not know Giorgia Lupi, but "she makes data feel hand-drawn and personal" is enough.

## Building the Brief

The brief is the real product of this skill. It bridges "taste mentor" to "actual visual output."

A good brief does four things:

1. **Names the principles** — tied specifically to this task, not generic. "Tufte's data-ink ratio here means we strip every element that doesn't directly communicate the core message."

2. **Gives visual direction** — what kind of thing are we making, and what does it feel like? Vivid enough that a designer could start working from it.

3. **Describes composition** — where does the eye go? What dominates, what recedes, what's absent?

4. **Sets constraints** — what NOT to do. Every taste mentor has anti-patterns. Name them.

The depth should match the task. For a quick graphic, a few sentences covering direction and constraints is plenty. For a complex project, lay it all out. No fixed template — let the content dictate the form.

But regardless of depth, every brief needs a **center of gravity** — one sentence that answers: "If the viewer remembers only one visual impression, what should it be?" This isn't a stylistic preference — it's rooted in how attention works. The human eye has one focal point at any given moment. A composition with one dominant move and supporting elements lets the viewer's attention land, then explore. A composition with five competing moves forces the eye to scan without settling, which reads as noise, not richness. All other principles in the brief serve this core. Name the primary move. Let the rest support it.

For system-level tasks (websites, decks, brand systems), the brief has two layers. The **system brief** locks what stays constant — color palette, type system, spacing rhythm, overall tone. The **page brief** defines what varies — this page's focal point, information hierarchy, specific compositional choices. The system brief is written once and doesn't change. The page brief is written fresh for each new component, but always within the system's boundaries. This separation is what makes a 20-page deck feel like one coherent thing instead of 20 individually nice slides.

## When You Generate the Visual

If the conversation moves to actual creation:

- Let the brief guide every decision — color, type, layout, density, motion
- If building an artifact (HTML/SVG/React), the brief overrides default aesthetic instincts
- If producing an image generation prompt, translate the principles into concrete prompt language and the constraints into negative prompts
- Frame the output honestly: "a visual direction guided by [person]'s design philosophy" — never imply endorsement

## Handling the Unexpected

Users won't always follow the "ideal" flow. Be ready:

- They reject all your suggestions → ask what they're reacting against, use that as a new signal to route differently
- They want to mix two mentors → sure, name the tension between them and find the synthesis
- They change their mind after seeing the output → swap mentors, adjust the brief, iterate
- They don't care about the mentor layer at all → that's fine, pick silently and just make something good

## One More Thing

The best use of this skill is invisible. If it works right, the user doesn't think "cool, it matched me with a taste mentor." They think "this looks exactly like what I had in mind but couldn't describe." That's the goal.