Bind audio.transcribe hydration to current message turn

src/daemon/routing.test.ts

@@ -656,7 +656,7 @@ describe('daemon command fast-path integration', () => {
 
     const keys = Array.from(router.agents.keys());
     expect(keys.some(key => key.includes(':research'))).toBe(true);
-    expect(processSpy).toHaveBeenCalledWith('compare k0s vs k3s for a homelab', undefined);
+    expect(processSpy).toHaveBeenCalledWith('compare k0s vs k3s for a homelab', undefined, undefined);
   });
 
   it('falls back to llm path when confidence is below fast threshold', async () => {
@@ -1938,6 +1938,6 @@ describe('daemon talk mode (voice wake) integration', () => {
       timestamp: Date.now(),
     } as MessageRouterInput, reply);
     expect(processSpy).toHaveBeenCalledOnce();
-    expect(processSpy).toHaveBeenCalledWith('what time is it?', undefined);
+    expect(processSpy).toHaveBeenCalledWith('what time is it?', undefined, undefined);
   });
 });

src/daemon/routing.ts

@@ -215,6 +215,26 @@ function persistLatestAudioAttachment(
   }
 }
 
+function extractLatestAudioToolInput(audioAttachments: Attachment[]): { data?: string; url?: string; mime_type?: string } | undefined {
+  const latest = [...audioAttachments].reverse().find((att) => (
+    (typeof att.data === 'string' && att.data.length > 0)
+      || (typeof att.url === 'string' && att.url.length > 0)
+  ));
+  if (!latest) {
+    return undefined;
+  }
+  const data = typeof latest.data === 'string' && latest.data.length > 0 ? latest.data : undefined;
+  const url = typeof latest.url === 'string' && latest.url.length > 0 ? latest.url : undefined;
+  if (!data && !url) {
+    return undefined;
+  }
+  return {
+    ...(data ? { data } : {}),
+    ...(url ? { url } : {}),
+    mime_type: latest.mimeType,
+  };
+}
+
 function isTtsEnabledForChannel(config: Config, channel: string): boolean {
   if (!config.tts?.enabled) {
     return false;
@@ -1317,6 +1337,7 @@ export function createMessageRouter(deps: {
       let messageText = incomingText;
       let attachments = msg.attachments;
       const audioAttachments = (msg.attachments ?? []).filter((a: Attachment) => isSupportedAudio(a));
+      const turnAudioToolInput = extractLatestAudioToolInput(audioAttachments);
       if (audioAttachments.length > 0) {
         persistLatestAudioAttachment(session, audioAttachments);
       }
@@ -1424,7 +1445,7 @@ export function createMessageRouter(deps: {
       let response: string;
       activeRuns.set(sessionIdForRun, agent);
       try {
-        response = await agent.process(messageText, attachments);
+        response = await agent.process(messageText, attachments, turnAudioToolInput);
       } catch (error) {
         const currentTier = agent.getModelTier();
         const canEscalate = deps.config.agents.auto_escalate && currentTier !== 'complex';
@@ -1434,7 +1455,7 @@ export function createMessageRouter(deps: {
 
         console.warn(`Auto-escalating session ${msg.channel}:${msg.senderId} from ${currentTier} to complex after processing failure.`);
         agent.setModelTier('complex');
-        response = await agent.process(messageText, attachments);
+        response = await agent.process(messageText, attachments, turnAudioToolInput);
       }
       const outboundAttachments = collector.drain();
       const ttsAttachment = await maybeBuildTtsAttachment(response, msg.channel);