Here’s the dilemma I keep running into when writing skill files — or vibe-coding them:
How do I know that a change I describe in plain English will actually produce the behavior I want, without introducing some unexpected side effect somewhere else?
It reminds me of something I experienced in college.
I first got BA in math & physics, then went on to study philosophy. My first year in philosophy was incredibly frustrating. There were so many competing ideas, often contradicting one another, and it wasn’t always clear which one was “right.”
Math and physics felt different. They were deterministic. Change an input, follow the rules, get the result.
Skill files increasingly feel more like philosophy than math.
So how do we test plain-English instructions with the same confidence we test code?