Struct TriggerExecutor 

pub struct TriggerExecutor { /* private fields */ }

Host-side trigger pipeline for one harness/session. Constructed by the CLI wiring with the parent agent + session handles and the same hook closures configured on the harness; handle_trigger replaces the old core entrypoint 1:1.

Implementations

impl TriggerExecutor

pub fn new( parent_agent: Arc<Agent>, parent_session: Session, runtime: TriggerRuntimeConfig, before_trigger: Option<BeforeTriggerHook>, on_trigger_prompt: Option<OnTriggerPromptHook>, before_trigger_action: Option<BeforeTriggerActionHook>, stream_fn: Option<StreamFn>, before_tool_call: Option<BeforeToolCallHook>, after_tool_call: Option<AfterToolCallHook>, ) -> Self

pub fn subscribe(&self, listener: TriggerListener) -> Box<dyn FnOnce() + Send>

Subscribe to the executor’s lifecycle events. Returns an unsubscribe handle.

pub fn abort(&self)

Cancel the in-flight trigger-prompt permission hook (if any). Mirrors the harness abort semantics for the old core-owned pipeline: the CLI wires Ctrl-C / /cancel through this alongside AgentHarness::abort.

pub async fn handle_trigger(&self, trigger: Trigger) -> EvaluationOutcome

pub fn notification_status_snapshot(&self) -> NotificationStatusSnapshot

Point-in-time view of the harness’s notification surface — the [TriggerRuntimeSnapshot] plus a Vec<NotificationHookStatus> collected from each registered hook via super::notification_hook::NotificationHook::status. The hook vec is a snapshot, not a live view; new registrations after this call are not reflected. Hook impls that have ended naturally still appear here until the next registration cycle — consumers should treat NotificationHookStatus.state as the source of truth for whether a hook is currently live.

pub fn abort_trigger(&self, trace_id: &str)

Cancel the in-flight sub-agent for trace_id. No-op if the trigger has already completed or was never accepted. The spawned task will observe the cancel inside its select!, abort the agent loop, and emit TriggerFailed with reason == "aborted" plus a trigger_result { success: false, summary: Some("aborted") } audit entry.

pub fn abort_all_triggers(&self)

Cancel every in-flight sub-agent. Each cancelled task writes its own trigger_result and emits TriggerFailed. Convenience wrapper around Self::abort_trigger for graceful shutdown.

pub fn register_notification_hook(self: &Arc<Self>, hook: DynNotificationHook)

Register a super::notification_hook::NotificationHook with the harness. Spawns two detached tokio tasks:

• Driver: calls hook.run(sink) and drives the hook’s transport (MCP read pump, cron watcher, etc.). Triggers the hook produces flow through the sink (an mpsc::UnboundedSender<Trigger>).
• Pump: reads from the sink’s receiver and calls Self::handle_trigger for each trigger. Exits naturally when the sender is dropped (e.g. when the hook’s run future ends).

The hook is stored for Self::notification_status_snapshot to read. There is no unregister API in this PR — hooks live until the harness is dropped or the driver task ends; the pump exits naturally when the sender closes. A later sub-PR may add explicit shutdown handles if a use case requires them; for now the YAGNI surface is “register and forget”.

self: &Arc<Self> because the pump task needs to clone the harness handle so handle_trigger is reachable from a 'static future. Callers already hold the harness as Arc<AgentHarness> in crates/harness::main so this is not a new ergonomic ask.