So, I'm writing a high velocity coding tool for programmers called Busy, that's going to be open source. It's a custom client for Claude that has coding super powers. Super interesting. Y'know what else is interesting? How I was totally wrong about the command scheme.
See, my first instinct was to parse the chat window like a scripting language. Invent a command interface where you could pattern match a lexical expression and pull out commands that way. But it's a very sloppy way to do things.
But what's cool is how I figured out that I could set up a command context that's basically a locked down powershell interface that Claude can use. Needs a couple of features like the lightest file editor ever written, but it's dog simple.
And it's so much less code.
I like the elegance of it. Tell the machine... "Hey, take a look at this piece of code, document everything in it, and write up some markup. Then, reverse engineer the thing from the spec." And it'll work.
Depends how complex it is.
But, not by a lot.
The method I came up with limits file size to under 6kb, which is perfect for microservice style architectures and clear separation of concerns within any given module.
I'm also writing a big cleanup for the metrics system I wrote over holiday break that drives the automated QA. Now, it's all going to be database driven, instead of JSON based, and should be accessible from the command context.
This is the fun stuff.