Explanation
Sessions that talk to each other, and how that differs from subagents
A spawned session is an ordinary session; a subagent runs inside its parent's own subprocess instead.
A session, mid-turn, can start two different kinds of extra work: another session, or a subagent. A spawned session comes from Turret's own tools; a subagent comes from Claude's own Agent tool. Both appear nested under the row of the session that started them, and neither is the other.
What a spawned session gets
A spawned session is not a special kind of session. It gets its own subprocess, its own sidebar row, and its own mailbox for messages sent to it. A session a person starts by hand gets exactly the same three things. It can be messaged, read, archived and stopped on its own, independent of the session that spawned it.
What a subagent does not get
A reader who has seen a spawned session get its own subprocess might expect a subagent to work the same way. It does not: a subagent runs inside its own parent's subprocess, with no subprocess, no sidebar row and no mailbox of its own. It cannot be sent a message or stopped by itself, because the CLI hands back a task id and no way to address it directly.
Two different limits on how deep either can go
A session a person starts may spawn a child, and that child may spawn a child of its own. The session at that second remove cannot spawn at all. A third level would add sessions nobody could realistically track, without widening what the chain can actually reach.
A subagent's own nesting inside a single subprocess is a separate limit entirely, and Turret does not cap it. A subagent costs no extra subprocess, so a subagent nested inside another spends no extra memory the way a spawned session's depth would.
What actually bounds a chain of spawned sessions
Turret tracks what each session's turns cost and never shows the figure anywhere on screen. What bounds a chain instead is the machine it runs on. Each live subprocess holds real memory, and a chain that grows past what the machine can hold slows everything else running on it.
What a spawned session's own instructions are, against a subagent's
A spawned session's own system prompt is built the same way as any other session's. It carries its mode's own prompt fragment, plus the preamble Turret adds to every session it starts. A subagent's guidance does not travel that route at all. Claude's own CLI builds a subagent's prompt, and Turret has no way to append to it directly.
Guidance a subagent needs instead travels through a hook Claude Code runs the moment a subagent starts. That hook carries extra context; it does not replace the prompt the CLI already built.
See Read a subagent's conversation for where to open what a subagent said and did. See Modes for the spawning capability every mode gives a session.