Skills, Commands, Sub-agents... It's All the Same Thing Now!
Claude Code used to give us a neat little taxonomy:
- Commands: quick, explicit “saved prompt” shortcuts (
/test→ runs a prompt). - Skills: richer procedural knowledge packs that the model could auto-discover based on intent.
- Sub-agents: “personas” that worked in a forked context so your main conversation stayed clean.
It was tidy. It was elegant. It was… doomed.