What does a subagent have of its own, rather than sharing it with the conversation that spawned it?

The principle — a subagent is defined by four things it does not share, and every useful property of delegation follows from that list.

Why the key is correct — Anthropic's definition is explicit: each subagent runs in its own context window with a custom system prompt, specific tool access, and independent permissions. The glossary repeats all four and adds that it works on a delegated task and returns a summary to the main conversation.

Why the others are wrong — the window is its own rather than a slice of yours. The system prompt and permissions are among the four independent properties, not shared ones. And a subagent is a separate assistant rather than a labelled section of your conversation.

Remember this — four independent properties: own window, own system prompt, its own tools, its own permissions.

Sources — Anthropic's Claude Code subagents documentation and glossary.