Diagnoses why a design reads as generated or template-built - scans the artifact against a 22-row anti-pattern table where every row names the tell, the perceptual mechanism behind it, and the skill that fixes it, then returns a severity-ranked, slug-routed fix list. Use when someone asks "why does this design feel AI-generated", "this looks like a template and I cannot say why", "what makes this page feel generic", "find the tells in this landing page", "why does this feel cheap even though it is clean", or work passes technical review yet reads mass-produced at a glance. Do NOT use for a general critique of usability and flow - use design-critique instead; to set up the design language that prevents the tells, use premium-design-foundations.
---
name: design-tell-inspector
description: Diagnoses why a design reads as generated or template-built - scans the artifact against a 22-row anti-pattern table where every row names the tell, the perceptual mechanism behind it, and the skill that fixes it, then returns a severity-ranked, slug-routed fix list. Use when someone asks "why does this design feel AI-generated", "this looks like a template and I cannot say why", "what makes this page feel generic", "find the tells in this landing page", "why does this feel cheap even though it is clean", or work passes technical review yet reads mass-produced at a glance. Do NOT use for a general critique of usability and flow - use design-critique instead; to set up the design language that prevents the tells, use premium-design-foundations.
---
# design-tell-inspector
Viewers cannot name why work feels generated, but they detect it in under 3 seconds - and every tell behind that verdict is mechanical, which means every tell is checkable. The clearest proof is the dark-glow default (near-black canvas, purple-to-blue gradient, untuned default grotesk): a look that once signaled craft and fully inverted into the signature of generated work, and tells keep inverting on roughly an 18-month cycle, so the table below is reviewed on that cycle. Without this scan, a team ships a page that passes every technical review and is dismissed at a glance.
## Operating procedure
Run the steps in order: inventory before scan, scan before ranking, ranking before routing. Severity depends on how many tells cluster, and routing depends on the completed scan, so score nothing until the whole table has run.
### Step 1: Gather the artifact and its charter
Ask for five inputs before scanning anything. If any input is a guess, label it a guess and move on.
1. **Artifact.** The page, screen, deck, image, or video frame itself - or its live URL and CSS. Never scan a verbal description: tells are visual and mechanical, and a description hides exactly the defaults this skill exists to catch.
2. **Audience.** Who the work is for. The more design-literate the audience, the harsher the scoring - a developer-tools buyer clocks the tells a general consumer glides past.
3. **Charter.** Paste the charter block if one exists. Without a charter, grade only the universal floors and the undifferentiated-cluster rows, and route charter creation to premium-design-foundations; charter-conditional rows (accent count and coverage, weight ceiling, canvas polarity, radius language) are skipped, never scored against house taste.
4. **Hunch.** Clarify what already felt off to the requester. Treat it as a search hint, not a verdict.
5. **Surface type.** Some rows apply only to web (scroll-jacking, performance), some only to imagery (physics contradictions, adjective-soup prompts). Skip inapplicable rows and say so.
### Step 2: Run the 22-row scan