Skill Creator

It enforces a three-level loading system where you front-load critical info, defer details to references folders, and design for context windows between 4k and 32k tokens. The approach is opinionated: imperative compression, progressive disclosure, lazy references. If you're writing skills that agents keep running out of context on, or you're annoyed at verbose prompt engineering practices, this gives you a strict framework. It's marked P0 critical, which feels a bit self-important for what's essentially a set of authoring guidelines, but the token economy angle is legitimately useful if you're shipping skills at scale.