Guard RTD websocket close for non-RTD sessions

deregister() now attempts to close rtdWebSocketManager whenever isSocketClosed is false, but RTD is optional and may never be initialized when aiFeature.realtimeTranscripts.enable is off. In that path, WebSocketManager still has the constructor placeholder (this.websocket = {}), so close() calls this.websocket.close() and throws, causing deregistration to fail for agents who never opened the RTD socket.

Useful? React with 👍 / 👎.

Addressed

Preserve default subscribe endpoint in websocket init

This refactor makes initWebSocket() depend on a caller-provided resource, but existing registration flow still calls initWebSocket({ body: ... }) for the primary socket in cc.connectWebsocket(). That means the main subscribe request can be issued with resource: undefined in runtime paths without strict TS enforcement, breaking the core websocket registration. Add a SUBSCRIBE_API default here or update all callers to pass a resource.

Useful? React with 👍 / 👎.

Addressed

Add reconnect wiring for RTD websocket manager

A second websocket manager is created for RTD, but only webSocketManager is wired into ConnectionService (the component that reacts to socketClose and re-initializes subscriptions). As a result, after network drops/keepalive timeouts, RTD can stay disconnected even if the primary websocket recovers, so transcript events silently stop until full re-registration.

Useful? React with 👍 / 👎.

Add reconnect wiring for RTD websocket manager

A second websocket manager is created for RTD, but only webSocketManager is wired into ConnectionService (the component that reacts to socketClose and re-initializes subscriptions). As a result, after network drops/keepalive timeouts, RTD can stay disconnected even if the primary websocket recovers, so transcript events silently stop until full re-registration.

Useful? React with 👍 / 👎.

check this if its valid. if yes lets have note for upcoming PR.

Preserve top-level type when re-emitting task events

This re-emit path now always uses payload.data.type, even though the same function still accepts task events via payload.data.type || payload.type when deciding whether to process them. For websocket payloads that carry task event names only at the top level (payload.type) and omit data.type, task.emit now fires with an undefined event name, so consumers listening for task lifecycle events will miss those notifications. The previous fallback behavior handled both shapes.

Useful? React with 👍 / 👎.

We have 2 different websocket connections so this will not be a conern

can we create a type for this and use it?

can u re-check this again we were doing below (payload.type is first priority)

const eventType = payload.type || payload.data.type;

Discussed it. This doesn't need to be addressed as this change was introduced in previous PR istelf

Add reconnect wiring for RTD websocket manager

A second websocket manager is created for RTD, but only webSocketManager is wired into ConnectionService (the component that reacts to socketClose and re-initializes subscriptions). As a result, after network drops/keepalive timeouts, RTD can stay disconnected even if the primary websocket recovers, so transcript events silently stop until full re-registration.

Useful? React with 👍 / 👎.

check this if its valid. if yes lets have note for upcoming PR.

Fail registration when RTD websocket setup fails

The RTD websocket initialization error is caught and only logged here, so register() still returns success even when the realtime subscribe/connect path failed. In that failure path (for example, auth/network errors on v1/realtime/subscribe), the SDK keeps the realtime transcript feature enabled but never receives transcript events and has no RTD reconnect wiring, which creates a silent partial-registration state for clients. Consider propagating this failure (or explicitly disabling realtime transcripts and surfacing a status event) instead of swallowing it.

Useful? React with 👍 / 👎.