Delegation as context management. What a subagent starts with, what it deliberately does not inherit, when Claude picks one on its own, and the difference between a subagent and a fork.

Two ways instructions persist without you retyping them: rules you scope to file paths so they cost nothing until they are relevant, and the notes Claude writes for itself between sessions.

The file that stops you re-explaining your project every session. Where it can live, how several of them combine, why it is context rather than configuration, and the test for whether an instruction belongs in it at all.