Inspiration
When I learn something new on roller skates, I watch a video first, standing still. Then I start rolling and half of it is gone. Was it knees bent? Where should I be looking? And for slalom I also need a row of cones that I never have with me.
So I thought: what if the glasses put the cones down for me and told me what to do, right when I need it, without making me look down at a phone?
What it does
You pick how many cones you want and how far apart they should be, and a ghost of the drill follows your gaze until you place it on the floor. On the start pad you get a quick reminder: eyes up, knees bent, arms forward. Then you go.
While you skate, the next cone glows brighter, and every now and then a short hint like "Bend knees" floats next to the path, a couple of cones before you need it. Only one at a time, and never in the middle of your view.
If you miss a cone, nothing yells at you. It just turns amber and you keep going. At the finish you see how it went, and if you cleared every cone the drill celebrates with a wave of light and neon rain. Then you hit Again, or change the drill and try something harder.
It works for quads and inlines alike.
How I built it
I started with the plan-lens skill. It interviewed me about the idea and turned my answers into a design brief with clear "done" criteria. Then CLAD agents in Claude Code built the Lens in Lens Studio through the MCP, and LEAF tests checked each criterion in the Preview.
Most of my work was looking at the Preview and writing design notes like "move this to the side, it covers the cones". A few tiny layout tweaks I did by hand.
Challenges I ran into
The biggest one was the narrow field of view. My first ideas looked fine on paper and broke in the glasses. Hints on the ground disappear when you look ahead, which is exactly what the drill teaches you to do. A menu in the middle covers the cones. Buttons near the floor make you stare at your feet. Almost every element moved at least twice.
Timing was the other one. A hint that pops up at the cone is too late to read, so hints now show up two cones early.
And some everyday stuff: Lens Studio sign-in didn't work in 5.24, so I switched to 5.23.2. Once the agent picked up the wrong project. And long agent runs eat tokens, so I learned to give smaller, sharper tasks.
Accomplishments that I'm proud of
The whole loop works without your hands while you're moving: set up, check your stance, ride, recover from a mistake, finish, go again. I'm also happy that the celebration happens out in the world and not in your face.
What I learned
On glasses, what's in view is the design. You can't just shrink a phone screen and float it in front of someone.
I also learned to be honest about what the Lens can't know. It can't see your knees, so it doesn't pretend to judge your stance. It just reminds you of the right one.
And the more precise the brief was (centimetres, degrees, seconds), the better the agents built it. The full story of what changed and why is in devpost/design-iterations.md.
What's next
Try it on real SPECS, on real skates. Add more drills like crossovers, T-stops and figure eights. Read the real ground and your speed, so hints adjust to how fast you're going.
Built With
- clad
- claude-code
- leaf
- lens-studio
- specs
- spectacles-interaction-kit
- typescript
