I turned a tiny WebGL arcade game into a browser-agent score challenge

I just shipped Llama Altiplaneta on Hugging Face Spaces.

On the surface it is a fast little browser arcade game: pilot a llama spaceship, dodge asteroids, survive, and chase a high score.

Important clarification: the game does not run a built-in AI model. The AI angle is that it can be used as an experimental browser-agent score challenge:

• playable directly in the browser
• persistent public leaderboard backed by a Hugging Face Dataset
• score target that humans and browser-control agents can both attempt
• machine-readable benchmark spec at /benchmark.json
• Open Graph / SEO previews so sharing the result actually looks decent

Try it here:

https://huggingface.co/spaces/thepowerofthepudu/Llama-Altiplaneta

Benchmark spec:

https://thepowerofthepudu-llama-altiplaneta.hf.space/benchmark.json

I am looking for feedback on three things:

1. Does the first minute feel fair?
2. Is the leaderboard enough to make you retry?
3. Would this be interesting as a tiny challenge for browser-control agents?

It is intentionally strange, a little chaotic, and very browser-native. If you try it, I would love to know your score and what broke first: the ship, the player, or the agent.