SKILL: brand-evaluation-rubric — Design Score / AI-Slop (Agents 03 and 08)

Purpose. Make principle P-03 ("nothing generic, with its own personality") measurable. It stops being opinion: two scores with an A–F grade and evidence.

Origin: adapted from gstack /plan-design-review. Used by Agent 03 (closing BRANDBOOK) and Agent 08 (at the QA gate, on the real app).

Two scores

• Design Score (A–F): does it have personality, hierarchy, coherence, intent?
• AI-Slop Score (A–F): how much does it look like "made by an AI with no taste"? (A = no slop; F = pure template.)

Report the at-a-glance reaction first (3 sentences): what it communicates, where the eye goes first, and a one-word verdict. That's the most valuable part; the rest is evidence.

AI-Slop detector (10 anti-patterns — each present lowers the grade)

1. Purple/blue hero gradient.
2. 3-column grid with icons in circles.
3. Inter (or another neutral sans) on EVERYTHING, with no second display font.
4. Everything centered, no asymmetry or focus.
5. Uniform, bubbly border-radius everywhere.
6. Floating blobs/decorations with no function.
7. Flat heading scale (36/24/18 with no drama).
8. Spacing with no rhythm (all 16/24, no real grid).
9. "Get Started"/CTA button identical to any SaaS.
10. Palette with no intent (>12 non-grey colors, or an accent with no usage rules).

Categories to grade (A–F each)

Typography · Spacing/rhythm · Hierarchy · Color · Interaction states · Motion · Density/focus · Coherence with the declared identity (BRANDBOOK/TOKENS) · Personality (memorable or template?) · Visual accessibility (contrast).

Output

Design Score: B+   |   AI-Slop Score: A-
Glance: "Feels like an esports scoreboard, not a SaaS. The eye goes to the win rate.
         In one word: competitive."
AI-Slop: 0/10 anti-patterns present (no gradient, no 3-col grid, 2 fonts...).
Top improvements: 1) ... (severity) 2) ...

• For Agent 03: if AI-Slop < A- or Design < B, iterate before closing the gate.
• For Agent 08: run the rubric on the real app (with qa-eyes), not on the mockup.
• Zero hardcoded hex/fonts outside the tokens = a requirement, not an opinion.

Rejection/approval criteria + workflow (merged from brand-evaluation, v6.3)

Folded in from the brand-evaluation skill (Agent 03) without losing content. Complements the scoring rubric with binary reject/approve criteria + Agent 03's workflow.

Reject internally and reformulate if it has one or more:

• Palette dominated by corporate blue + white + grey.
• Generic sans-serif typeface with no defined personality.
• Dashboard layout (left sidebar + header + cards) as the main structure with no specific justification.
• Landing page with a big hero + three feature columns.
• Trendy gradients unrelated to the product's personality.
• Generic line iconography with no system of its own.
• Anything that could serve any other product without changing a thing.

Approve if:

• The palette has an explainable narrative rationale.
• The typography has hierarchy with purpose.
• The layout responds to how the user uses the product.
• Someone can tell what kind of product it is without reading the name.
• The proposal could not be applied to another product without modifying it.

Workflow (Agent 03):

1. Gather references from the PO before proposing anything.
2. Generate at least 3 distinct visual directions.
3. Internal evaluation against the rejection criteria.
4. Present to the PO with enough context to decide.
5. Iterate until explicit approval.
6. Convert the approved brandbook into CSS tokens.

Common mistakes: proposing before gathering references · presenting variations of the same concept as distinct options · moving to tokens without explicit approval · defining tokens by size instead of by use.