fix: isolate MCU background task results (#2181)

Author: EKKOLearnAI

docs/chat-chain-changes/2026-07-22-mcu-background-result-speech.md

@@ -0,0 +1,24 @@
+---
+date: 2026-07-22
+pr: pending
+feature: MCU background result speech
+impact: MCU voice turns return to idle after the parent response and later speak only the final autonomous Agent response produced for completed background work.
+---
+
+The Global Agent server and the direct outbound MCU relay compatibility path
+keep a hidden session listener while a voice-triggered Hermes run has pending
+background delegations. Subagent telemetry and delegation lifecycle events
+remain server-internal and are not forwarded to the MCU. When the background
+completion starts its autonomous parent turn, the listener buffers that turn
+and reuses the existing MCU audio queue only after `run.completed` provides the
+final Agent output. Existing MCU firmware and the transparent remote relay
+server require no protocol changes. An MCU interrupt received after the parent
+turn is already idle no longer records a deferred session abort, so immediately
+starting another recording leaves pending background delegations running; an
+interrupt during an active foreground turn still aborts that turn normally.
+Each MCU foreground run now carries an internal `queue_id`. When a new voice
+turn is queued while an autonomous background delivery is returning on the
+same chat session, MCU listeners use that identifier to keep the two event
+streams separate. The existing background listener owns the autonomous result;
+the new listener ignores it until its own queued run starts. This identifier is
+not part of the MCU wire protocol, so older firmware remains compatible.

packages/server/src/services/global-agent/outbound-relay-client.ts

@@ -412,7 +412,7 @@ export async function handleBridgeRun(
   // boundaries and remain disabled independently of this default.
   const backgroundDelegationEnabled = data.background_delegation_enabled !== false
   if (!session_id) {
-    socket.emit('run.failed', { event: 'run.failed', error: 'session_id is required for cli source' })
+    socket.emit('run.failed', { event: 'run.failed', queue_id: data.queue_id, error: 'session_id is required for cli source' })
     return
   }
 
@@ -660,13 +660,15 @@ export async function handleBridgeRun(
     pushState(sessionMap, session_id, 'run.started', {
       event: 'run.started',
       run_id: started.run_id,
+      queue_id: data.queue_id,
       queue_length: state.queue.length || 0,
       autonomous: data.autonomous === true,
       delegation_id: data.background_delegation_id,
     })
     emit('run.started', {
       event: 'run.started',
       run_id: started.run_id,
+      queue_id: data.queue_id,
       queue_length: state.queue.length || 0,
       autonomous: data.autonomous === true,
       delegation_id: data.background_delegation_id,
@@ -695,7 +697,7 @@ export async function handleBridgeRun(
         currentInputTokens,
         shouldPersistUserMessage && displayRole === 'user',
         data.model_groups,
-        { autonomous: data.autonomous === true, delegationId: data.background_delegation_id },
+        { autonomous: data.autonomous === true, delegationId: data.background_delegation_id, queueId: data.queue_id },
       )
       if (chunk.done) {
         sawTerminalChunk = true
@@ -741,7 +743,7 @@ export async function handleBridgeRun(
         currentInputTokens,
         shouldPersistUserMessage && displayRole === 'user',
         data.model_groups,
-        { autonomous: data.autonomous === true, delegationId: data.background_delegation_id },
+        { autonomous: data.autonomous === true, delegationId: data.background_delegation_id, queueId: data.queue_id },
       )
     }
   } catch (err: any) {
@@ -791,6 +793,7 @@ export async function handleBridgeRun(
       background_pending: backgroundPendingCount(state),
       autonomous: data.autonomous === true,
       delegation_id: data.background_delegation_id,
+      queue_id: data.queue_id,
     })
     if (queueLen > 0) {
       dequeueNextQueuedRun(socket, session_id)
@@ -1125,7 +1128,7 @@ async function applyBridgeChunkAsync(
   currentInputTokens = 0,
   currentInputIncludedInDb = true,
   modelGroups?: RunModelGroup[],
-  runMetadata?: { autonomous?: boolean; delegationId?: string },
+  runMetadata?: { autonomous?: boolean; delegationId?: string; queueId?: string },
 ): Promise<void> {
   if (state.activeRunMarker !== runMarker) {
     bridgeLogger.info({
@@ -1653,6 +1656,7 @@ async function applyBridgeChunkAsync(
     background_pending: backgroundPendingCount(state),
     autonomous: runMetadata?.autonomous === true,
     delegation_id: runMetadata?.delegationId,
+    queue_id: runMetadata?.queueId,
     workspace_run_change: workspaceRunChange,
   }
   emit(eventName, payload)

packages/server/src/services/global-agent/server.ts

@@ -237,6 +237,7 @@ export class ChatRunSocket {
         socket.emit('run.failed', {
           event: 'run.failed',
           session_id: data.session_id,
+          queue_id: data.queue_id,
           error: err instanceof Error ? err.message : String(err),
         })
         return
@@ -331,6 +332,7 @@ export class ChatRunSocket {
         socket.emit('run.failed', {
           event: 'run.failed',
           session_id: data.session_id,
+          queue_id: data.queue_id,
           error: err instanceof Error ? err.message : String(err),
         })
       }
@@ -479,11 +481,13 @@ export class ChatRunSocket {
         const payload: {
           event: 'run.failed'
           session_id?: string
+          queue_id?: string
           error: string
           queue_remaining?: number
         } = {
           event: 'run.failed',
           session_id: data.session_id,
+          queue_id: data.queue_id,
           error: `Agent Bridge is not reachable: ${bridgeReady.error}`,
         }
         if (queueRemaining > 0) payload.queue_remaining = queueRemaining

packages/server/src/services/hermes/run-chat/handle-bridge-run.ts

@@ -379,6 +379,7 @@ describe('ChatRunSocket bridge readiness gating', () => {
     expect(socket.emit).toHaveBeenCalledWith('run.failed', {
       event: 'run.failed',
       session_id: 'session-1',
+      queue_id: 'queue-failed',
       error: 'Agent Bridge is not reachable: bridge offline',
       queue_remaining: 1,
     })