A toolkit serves every caller from one artifact. You write it once; the whole system can use it.

• An agent reads the toolkit's skills and calls a command over bash — run-command <name>.

• A workbook's HTML calls it live through the Dock (the closed loop: a button that runs the thing, not a "copy as prompt").

• A workflow orchestrates it as one step.

• Another toolkit calls it via the commands capability — composition.

• The fabric loops a kernel over many inputs at once.

Discovery is one query: a toolkit is a :toolkit:-tagged node, found by (tags :toolkit:). An agent sees a compact index and reads a skill only when it needs it.

• How agents see toolkits: Authoring skills

• Distribute compute: The fabric