If you mostly use chat-style AI and keep trying to follow every new AI tool, this is the expensive mistake: treating every update like a model story. You see a repo update, almost scroll past, then wonder whether you are missing the one detail that should change your next move.
My read from addyosmani / agent-skills is simple: CLAUDE.md beats a model upgrade when team habits still live only in people's heads. If it is not written in the rules file, it does not exist for the AI.[C002]
Why: the public docs put rules files in layer 1, the top context layer, and name CLAUDE.md and AGENTS.md as the place to store team habits. That means the AI sees written rules before the rest, not after. The same docs also reject the idea that an agent will just pick up your norms on its own.[C001]
That is the real cost of chasing shiny upgrades first. You can spend time, budget, and attention comparing models while the most reusable part of your team knowledge is still invisible to the tool. A tool update is not worth your time because it lists more features. It is worth your time if it changes your next decision.
Boundary: this is a docs-only read of the public repo, not a benchmark on one model, IDE, OS, or live codebase. Next move: write the rules file first, then compare models. Share this with the person pricing upgrades.