Hexmap: Week 0x23

The artificial interns really came through this week! I was revisiting the hex detail page and got triggered to work on a stretch feature: mapped hexes will now be “enriched” with relevant Wikipedia entries. I’m surprised there are so many discovered, though may have to allow being more aggresive with the Wikipedia APIs to keep up.

Now, this isn’t that complicated of a feature. But I really matured the background process and remote service endpoint management as part of it. All while on trains between Amsterdam and Berlin. I’ve never been great at finding flow state during travel, but using Kilo Code lets me be at least as productive without it. (I have very mixed feelings about this.)

Speaking of the LLMs, here’s a noteworthy snippet from a conversation with them about a mistake they’d made. I think it well documents the current state of my workflow with them:

You’re absolutely right - this was a conceptual confusion that could have been prevented with clearer documentation. Let me review the memory bank files to identify where we can clarify the distinction… Looking at the memory bank, I can see that while we have good definitions of “Hex” and “Visit” in the conceptual model, the data model documentation could be much clearer about this critical distinction.

This being my last week before starting a new job, I got a little burst of “wrap it up” panic and put back together a lot of messy layouts into good-enough state. Then I deployed a user feedback tool so I can invite some private beta users. Doing that right meant I needed OAuth, which I leaned heavily on Claude Sonnet 3.5 to implement. Also for the first time I asked it to build a UI (for administration of OAuth clients) and let it self-test through browser control.