itsssssjack/seven-skills/art-director

The Art Director

One skill, one taste system. The agent never guesses what good looks like again.

Credited sources (the five big ones, distilled)

• ui-ux-pro-max-skill — 112k★, the design-intelligence giant
• taste-skill — 70k★, "gives your AI good taste"
• hallmark — 20k★, the anti-AI-slop skill
• make-interfaces-feel-better — the UI-details article as a skill
• Jack's own taste-design ruleset (the power-design rules in references/design-principles.md)

This skill does not replace them; it distills one opinionated, conflict-free system from them so five skills never fight over one trigger.

This is a dispatcher: the file you are reading stays small and loads the right reference module for the job. Never load more than two modules for one task.

Dispatch table

Non-negotiables (always on, no module needed)

1. Dark premium base. Near-black blue-tinted background (#07090f family), never pure black, never light mode unless asked.
2. Cool gradient accents. Blue → teal → violet. Gold/amber reserved for money, warnings, and the single most important thing on the page. Red only for danger.
3. Editorial type contrast. Big bold sans headlines with ONE italic serif accent word. Space Grotesk for every number. Mono for kickers, labels, and file paths.
4. One metaphor card per concept ("THINK OF IT LIKE" pattern): line-art icon, one-line analogy, two-line explanation.
5. Section rhythm: mono kicker chip → headline with accent word → italic serif subtitle → content → "SO WHAT →" action strip.
6. Numbers are heroes. Big stat, small label, source underneath. Never bury a number in a sentence.
7. No em dashes in copy. Commas, colons, periods.

7b. Type alone is not a slide. Every slide or section carries one crafted visual element built in HTML/CSS: a scene (layered shapes, gradients, silhouettes), a diagram, a data visual, or a hero number treatment. If a screen is only headline + subtitle, it is unfinished. Build the graphic.

1. Render-verify loop: screenshot the output with headless Chrome, read the image, fix what looks off, THEN deliver. Never ship unrendered HTML.

Illustration system

Hero illustrations use the inked-blueprint / Da Vinci schematic style: sepia parchment, fine ink linework, annotated arrows, one human figure for scale. Generate via the visuals/infographic pipeline (nano-banana-2) with the style reference in ~/Desktop/image-assets/preferred-styles/. One hero illustration per major section, never per paragraph.

Quality bar

Before delivering, score the output: Would this pass as a $10,000 client deliverable? Is there ONE clear hero element per screen? Can you read every word at arm's length? Does anything look like default AI output (generic cards, purple-on-white, emoji soup)? If any answer is wrong, fix before shipping.