Inspiration
Fingerboards are fun because you can pull one out almost anywhere and have a quick skate session. I wanted to bring that to mixed reality: wherever you are, with or without a flat surface in front of you, you can build a park without buying physical obstacles and get a quick session in.
What it does
Glitch Skate! is a fingerboard you ride with your bare hands in mixed reality, with no controllers. Using your index and middle fingers directly on the board, you can perform tricks like ollies, kickflips, heelflips, shove-its, grinds and manuals. Tricks come directly from how you move your fingers on the board.
- Build your own park on a virtual table or a real surface in your room, place obstacles, and save your parks to come back to.
- Skate lines: unlike real fingerboarding, you're scored on chaining tricks into combos.
- Take on challenges: short, optional levels that unlock new obstacles for the park builder. Free skating is always available.
How I built it
I built Glitch Skate! in Unity with the Meta XR SDK: hand tracking for all input, passthrough so you skate in your own room, and the MR Utility Kit to scan the room so the board can ride on your real surfaces. The board is a physics object driven by your fingertips, and tricks are detected from how your fingers move.
I used an AI-assisted workflow with Claude Code and the official Unity plugin.
I also used ElevenLabs for sound effects and music, and ChatGPT to create 2D textures and animated text.
It started as a trick-sandbox prototype in September 2026 (video). Everything else (park building, scoring, challenges, additional tricks and tuning, onboarding) was built during the competition.
Challenges I ran into
Hand gesture detection was hard to balance, because the same motion can mean different things. Am I moving my index finger in midair to do a kickflip, or do I want to rotate the board?
Early on, I used a spec-driven workflow: a Markdown file (written with Claude's help) describing each mechanic in detail, which I iterated on as I tested different approaches. Later, the bottleneck became the time spent hand-tuning gesture properties in the Unity editor. So I changed the workflow: while I play, the game records its state every frame, and I press a key whenever something goes wrong to label that moment. Claude then analyzes the recorded data and adjusts the gesture-detection parameters based on the labeled failures.
Accomplishments that I'm proud of
Gesture detection was really unreliable at the start. It's now at the point where I can reliably land the trick I'm going for, which makes the fingerboard feel much more natural to control.
I'm also proud of how far the project came during the competition. What started as a trick sandbox now has a park builder, saved parks, combo scoring, challenges, unlockable obstacles, and onboarding. It feels much closer to a complete game than the original prototype.
What I learned
I learned that AI-assisted development works especially well when the agent has access to concrete gameplay data instead of relying entirely on written descriptions of problems.
Pairing subjective feedback from playing the game with recorded state data made gesture tuning much faster and more reliable than adjusting values manually in the Unity editor.
What's next for Glitch Skate!
I'm planning to release Glitch Skate! in Early Access in December 2026 and start responding to real player feedback.
On the roadmap:
- A community park browser
- More challenge levels and obstacles for the park builder
- More tricks
