Visual Identity

Overview

A deep taste interview that produces a concrete, non-generic visual direction: narrative rationale plus implementable design tokens. The explicit goal is to avoid the default "AI app" look (safe sans-serif, blue/purple gradients, uniform rounded corners) by grounding choices in real references and comparisons rather than the interviewer's own defaults.

When to Use

• Before or during frontend implementation, when visual direction isn't set
• User wants the project to feel "designed", not templated
• Follows idea-interrogator in a validation → design → brief pipeline, but works standalone too

Process: Deep Taste Interview

Run multiple rounds, one question at a time — wait for the user's actual answer before asking the next question or proposing any direction. Don't answer your own questions and don't shortcut to a single questionnaire; the point is triangulating taste through back-and-forth comparison, not collecting a spec sheet in one pass.

Round 1 — References: Ask for 2-3 things (apps, brands, physical objects, eras, films) whose look the user admires, and what specifically about each.

Round 2 — Mood words: Ask for 3-5 adjectives describing the desired feeling (e.g. "quiet and confident" vs "loud and playful"). Push for specificity — "modern" and "clean" are non-answers, ask what modern actually looks like to them.

Round 3 — Comparisons: Present concrete forced-choice pairs based on rounds 1-2 (e.g. "serif or grotesque sans for headings?", "high contrast dark mode or warm paper-toned light mode?", "sharp corners or soft?"). Use at least 4 comparisons covering: typography style, color temperature/ contrast, corner/border treatment, density (spacious vs. dense).

Round 4 — Practical constraints: fonts (if any hard requirement), required brand colors (if any), theme (light/dark/both), and purpose/audience (reuse from idea-interrogator-verdict.md if present in cwd instead of re-asking).

Only after all four rounds have real answers, synthesize them into a direction. If answers conflict, name the tension and pick a resolution rather than averaging into blandness.

Deliverable

Write visual-identity.md to the current working directory:

# Visual Identity: <project name>

## Direction
<3-6 sentence narrative: the chosen aesthetic and why it fits the project's
purpose/feeling/references, naming the specific reference points used>

## Tokens

**Typography**
- Heading font: <name>, stack: `<font stack>`
- Body font: <name>, stack: `<font stack>`

**Color**
- Background: `#hex`
- Surface: `#hex`
- Text primary: `#hex`
- Text secondary: `#hex`
- Accent: `#hex`
- (dark-mode variants if applicable, same keys)

**Shape & Spacing**
- Corner radius: `<value>`
- Border style: `<value>`
- Spacing scale: `<e.g. 4/8/16/24/32/48px>`

**Theme**
- Mode(s): light | dark | both

Common Mistakes

• Answering your own questions instead of waiting for the user's real answer — this collapses the interview into a monologue and defeats the point of a taste-triangulation process.
• Jumping straight to token values without the reference/mood/comparison rounds — this is exactly how generic output happens.
• Picking the "safe middle" on forced comparisons instead of committing to a point of view.
• Descriptive-only or tokens-only output — the deliverable requires both.