Building the Arena: How I Used Super Skills to Rebuild My Game System

I found this plugin on Claude today called Super Skills, and I used it to re-engineer the game arena project I've been working on. It solved a bunch of problems I kept running into.

The core issue was always the same: planning at scale doesn't work the way I wanted it to. I'd flip Claude into planning mode, we'd go back and forth on ideas, break it into phases, and by the end I'd have this massive planning document. Claude would ask, "Do you want me to execute this?" And the answer was always no, because the plan was huge. Too huge. It wouldn't fit in a context window. The work would get fragmented. Nothing got built.

So I changed my approach. I opened a ChatGPT 5.2 session with deep thinking enabled, and I worked through the high-level design in chunks. Game engine. Identity and ranking. Leaderboard mechanics. Terminal UI. One piece at a time, thinking deeply about each one.

And this is where it got interesting. When I got to the ranking system, I realised something I wouldn't have figured out naively. You can't just use points to rank players. There's this whole mechanism to it. And you can't rank them in one place: you've got local rankings and tournament rankings. That's a distinction that matters. Working through it with ChatGPT forced me to understand why.

For each segment, I'd ask the model to produce a markdown design document. Then I'd take that, clean it up, stick it in the docs folder in my repo. By the end, I had a cohesive picture of the whole system, even down to the terminal UI using the Charm library. I asked ChatGPT to mock it up in ASCII, and they were great. Simple way to visualise what it should actually look like.

So my workflow was: ChatGPT for deep thinking on design, producing persistent design docs. Then I moved to Claude Code with the Super Skills plugin and said, "Look at the overview, follow the docs it links to, build me an execution plan."

That's where things got really impressive.

Super Skills broke down the work properly. It understood the phases. It made a table of tasks. But the thing that really got me was this: the plan saved itself into a markdown file in the repo. Not in the chat. In the actual project. Persistent. Not deleted when the session ends.

I've spent so much time telling Claude, "Okay, you've written this plan, now save it to the repo so I can work through it chunk by chunk." That's been one of my biggest complaints about the context window problem: not having the ability to partition work properly, save it, and come back to it. But Super Skills just does that automatically. And it knows about agent swarms. It farms the work out so new context windows can handle it. It partitions the work asynchronously.

I rebuilt my entire system in one day. Switched languages from TypeScript to Go, completely new architecture, fresh start to finish. That's a story for another time. But the point is, with Super Skills managing the workflow, breaking it down, farming it out, keeping the code quality tight through loads of tests.

That's the thing about building at scale. Context windows are finite. You can't work around that. But if you save your plans, keep them persistent in the repo, break the work into pieces small enough to execute in isolation, and let agents work asynchronously without wasting context on task switching, you can build something real in a day.

Super Skills does all that automatically. And I was really impressed by it.