I've been playing around with a new DSL, and I've had good luck adding a "describe" feature, eg,

  newdsl describe some_feature

This returns the docs for that language feature. So instead of a giant SKILL.md containing the language spec, it exposes a way for an LLM to introspect.

DSLs climb the ladder of abstraction and constrain the solution space at the language level, which IMO provides tighter feedback to both humans and machines.