Claude Skills
A concept product for persistent AI instruction sets — tell Claude once, it applies your preferences forever.
The problem
Every Claude conversation starts from scratch. You re-explain your tone, your context, your preferences — session after session. The model is capable; the interface forgets.
My hypothesis
If you frame AI customization as 'skills' rather than 'system prompts,' you remove the technical barrier. Most people don't know what a system prompt is. Everyone knows what a skill is.
What I built
A concept site that shows the product working before asking anyone to sign up. The core bet: lead with a live chat preview demonstrating a skill in action, not a feature list. Two pathways — browse pre-built skills or create your own — sized for two different mental models.
What broke
The site shows what persistent skills could feel like — but the actual Claude interface doesn't support this natively. I was designing for a gap in the product, not an existing surface. The thing I shipped is a pitch, not a product.
What I learned
Building a concept product is different from building a functional one. The constraint isn't 'can I build this' — it's 'can I make someone believe this should exist.' That's mostly a copywriting and sequencing problem.
If I kept going
If Anthropic ships persistent memory or a skills layer, this design is a ready-made reference. The real test would be a working prototype — a Chrome extension that actually persists instructions across sessions would turn a concept into a proof of concept.
What was built
Design
UI/UX concept and information architecture
- Landing page layout
- Live demo component
- Use case personas
- CTA hierarchy
Frontend
Static site built with vibecoding
- Next.js
- Tailwind CSS
- Vercel deployment
Prototype
Landing: tell Claude once, it remembers forever
How it works: pick a skill, Claude opens, press Enter
Browse the gallery: 9 pre-built skills ready in 30 seconds