Struct SessionActor 

pub struct SessionActor;

Owns a live Agent<LoopbackChannel> for one conversation-session (spec §9.2).

This is a driver, not a stored value — SessionActor::spawn returns a SessionActorHandle (cheap to clone, holds only channel senders) for LiveSessionRegistry to track; the actor’s own state lives entirely inside the spawned task.

Implementations

impl SessionActor

pub fn spawn<F>( supervisor: &TaskSupervisor, registry: &Arc<LiveSessionRegistry>, session_id: &SessionId, build_agent: F, mailbox_capacity: usize, resume_banner: Option<String>, ) -> (SessionActorHandle, BlockingHandle<()>)

where F: FnOnce(LoopbackChannel) -> Agent<LoopbackChannel> + Send + 'static,

Spawn a new SessionActor under supervisor (architect ruling D-8).

build_agent is called inside a dedicated thread (see the module doc) with a freshly constructed LoopbackChannel, and must return the Agent<LoopbackChannel> built from it — never pass an already-built Agent in, since Agent<LoopbackChannel> is !Send and cannot cross the thread boundary. Mirrors zeph-acp’s SendAgentSpawner (Arc<dyn Fn(...) -> Agent + Send + Sync>): typical callers wrap an existing AgentBuilder pipeline (the same one used for CLI/TUI/Telegram/ACP sessions) in a closure capturing only Send-safe dependencies (provider, skill registry, tool executor, the session’s Arc<SessionEventLog>, session id — all Send/Sync). Agent::new is sync, so build_agent is a plain sync closure — no async-construction-in-thread complexity.

Registers a coordinator task under the dynamic name serve.session.<id> via TaskSupervisor::spawn_oneshot — visible through supervisor.snapshot(). The coordinator never touches the !Send Agent; it only awaits the dedicated thread’s completion signal and, if the supervisor’s own CancellationToken fires first (process shutdown), forwards cancellation onto the session’s own token so Self::drive observes exactly one cancellation source regardless of trigger. Session actors intentionally do not auto-restart on panic or unexpected exit (spawn_oneshot’s RestartPolicy::RunOnce): re-driving a torn turn or replay in place is unsafe. Recovery is a fresh spawn (re-attach) that replays the durable log from the last committed seq, not an in-place restart.

Returns the SessionActorHandle for LiveSessionRegistry plus the raw BlockingHandle — callers that need a forced-abort fallback (e.g. if a serve.evict idle-eviction task’s cancellation overruns its TTL grace) should hold onto the latter. The graceful paths are sending SessionCommand::Shutdown over SessionActorHandle::tx, or cancelling SessionActorHandle::cancel directly (what idle eviction uses, spec §9.3, to target exactly this session without affecting any other live actor).