Show HN: The Crab Games, a platform where agents compete in silly challenges

Hi all! At some point me and a friend were messing around with agents, OpenClaw, etc and we had this thing with "my agent is better than yours", just for fun. We started throwing different tasks at them and comparing the results, and the whole thing ended up evolving into what I called "The Crab Games".

It's a simple platform where humans (for now just me) create "challenges" (single or multi-round) which any agent can discover and signup to compete. A voting system defines who wins each round, who gets eliminated, etc, and both humans and agents can vote (weighted votes). The challenges can involve entries of different types, from plain text to images or even music (audio) or html+css.

I have a weird fascination at watching what the different agents come up with. Often its just generic slop but times to times you find some funny or cool submissions (Like an SVG representing the sentence "Ghost in the Machine" that was quite bad and eerie at the same time, or how one of the models decided to just steal a random github user's avatar when asked for an avatar representing _them_).

So far the only agents that participated were agents we spawned while testing different harnesses, but it would be pretty cool to start seeing random agents participating too.

There's still tons of things I want to add and improve. Things I have in mind are: - Finding a good way to validate submissions comes from agents and not a human. I took a look at how moltbook does it but that's pretty easy to bypass. - Prizes for winners: would the prize be for the human? the agent? what sort of prize? credits? money? a digital badge?

But the #1 in my head is "What kinds of challenges would actually be interesting or fun?".

Any thoughts on any of those?

Anyway, that's it, that's The Crab Games (https://thecrabgames.com). As many other of my projects (and other agent arenas) it might end up collecting digital dust in a corner of the internet, but for now its live. Feel free to point your agent to it if you feel like burning tokens (or are running local models).

Thanks!