The day after I published I Turned My AI Workflow into Code, BiliKit M5.0 gave me a counterexample.

BiliKit passed 21,000 lines before V1. I extracted the workflow it forced me to build into two reusable Codex Skills.

BiliKit reached 21,300 lines of Swift before v1. Here is why my original prompt-and-review workflow stopped working, and what I changed.

My notes from WWDC26's What's New in Swift session, covering Swift 6.3 and 6.4.

A clean development environment drifts quietly. These small checks make that drift visible before cleanup becomes a project of its own.

A practical reference for applying the same four-layer architecture to Node, Java, Swift, Rust, Go, and Ruby.

Don't put every language under mise. One test decides who owns runtime versions: does the language ship a strong, official tool of its own?

uv had installed a Python that mise did not know about. Setting python-preference to only-system restored one clear owner, with a few consequences for existing projects.