js13kgames is a long-running game jam with a simple goal: create a JavaScript game in 13 Kilobytes of zipped JavaScript. I’ve participated 8 times since 2017. They also send a t-shirt to the top 100 entries, which has been a good part of my wardrobe since 2017.

I was unsure of joining this year due to the usual lack of time, but decided to continue experimenting and evaluating the current extent of LLM’s powers, and improving my AI-powered development flows; my “entry” last year followed the same line of thought, and it was not very good, but neither were the LLM tools nor my skill of using them, so here we are again.

These two last entries have been something of a change on how I perceive js13kgames; the first years it had a big element of coding craftsmanship to manage to fit as much as possible within the limits; hard design decisions were always there, but they were layered on top of that; while some participants still continue on this same path (and, I reckon, for some of them this is even a good opportunity to code something on their own), to me at least this year it’s going to be mostly about optimizing design and higher level technical decisions and see what I can manage to build (at least that’s the plan, we’ll see how I end up trying to manually optimize the code and curse claude the final days to fit a much needed feature).

As an aside, I’ve been recently thinking of making a bigger scale grand strategy game set in the Colombian Civil Wars of the XX turn-of-century, so I thought it would be good to have a smaller project to practice the overall design of such a game. I’m a fan of Paradox grand strategy games, mainly of the concept behind them, even if I haven’t been hooked by their gameplay. So my idea has some elements from them, but with some important changes especially in the territory model (interconnected nodes at the forefront instead of “provinces” with an area).

So, I created an initial prompt reflecting this design, after some time it landed in a pretty decent already working prototype, well within the size limits of the compo.

The first rounds of feedback were the most critical that claude couldn’t get from its execution; the game was too quick, and the movement of the units wasn’t smooth. Also some critical features, like armies fighting each other while moving, were missing.

The process after this was playing the game, trying to finish a complete session, and provide feedback to improve the gameplay and the user experience. From the onset claude set up a test harness where it simulated hundreds of battles and moved the numbers to make sure the games were winnable, which was a great feature.

The immediate improvements were adding a cost to fleeing to battles and making building armies and repairing defenses take some time, as well as allowing splitting armies, and traveling to far away nodes with pathfinding,

Further improvements focused on usability tweaks, but also adding difficulty levels since the game was too easy; these focused on artificial boosts to stats instead of increased intelligence, since we needed something quick and easy.

Another big addition was the fog of war, making it so that the player can’t see details of what’s going on in far away territories (unless they send a scout). This didn’t affect AI at all, thus also giving it a bigger advantage

And so ended day 1 of development – it’s working, but there’s little unique to it, and visuals could be improved. So that’s what’s coming next.