So my automated QA system is a massive pile of diverging features and JSON files. Time to consolidate and do this a little more systematically. So I'm putting everything into a SQL database. All purdy like. Here, you can see my happy little trending views.
Basically, what this does is track the cognitive complexity and basic structure of your code base. It's implemented as a git hook, and both the user and the AI agent can pull this information, more or less, in real time to get a picture of system health for the project, module, or specific file they want to work on.
Cognitive complexity is a metric that tracks how easy your codebase is for a semi-autonomous AI agent to understand and edit. They don't have a lot of cognitive bandwidth, so you have to think about that when you're using them on bigger code bases for high velocity projects. Otherwise, man oh man, are you in trouble, lol.
I figured, the best way to handle this was by building it into the command set. So the agent can call up a file or decide to look at the metrics for a model, and a) get a better idea of how the project is constructed, and what they're getting into as far as editing feasibility. We have four levels of complexity scoring simple, medium, high, and DANGER.
Doing it this way solves the problem of detective work and wild assumptions on the part of the agent. It enhances clarity and reduces hallucinations in your code.
Also, it's going to make the gamified elements of the system that much more fun to play.