Subagents

I used to write Agent Skills as flat instruction files. This basically meant that a set of repeatable instructions I had would be converted into one skill, and before too long, I would have around 50 skills per repo. Each skill described one workflow, and they never called each other. That left a lot of repeated text across them.