bloby-bot 0.40.0 → 0.41.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bloby-bot",
-  "version": "0.40.0",
+  "version": "0.41.0",
   "releaseNotes": [
     "1. # voice note (PTT bubble)",
     "2. # audio file + caption",

package/supervisor/channels/manager.ts

@@ -116,6 +116,7 @@ export class ChannelManager {
       },
       (status) => this.handleStatusChange(status),
       (audioBase64) => this.transcribeAudio(audioBase64),
+      (fromMe, isSelfChat, isGroup) => this.shouldProcessWhatsAppAudio(fromMe, isSelfChat, isGroup),
     );
     this.providers.set('whatsapp', whatsapp);
 
@@ -140,6 +141,7 @@ export class ChannelManager {
         },
         (status) => this.handleStatusChange(status),
         (audioBase64) => this.transcribeAudio(audioBase64),
+        (fromMe, isSelfChat, isGroup) => this.shouldProcessWhatsAppAudio(fromMe, isSelfChat, isGroup),
       );
       this.providers.set('whatsapp', whatsapp);
       provider = whatsapp;
@@ -419,6 +421,35 @@ export class ChannelManager {
     return config.channels?.[channel];
   }
 
+  /** Decide whether an inbound WhatsApp audio is worth transcribing.
+   *  Mirrors the gates in handleInboundMessage so we don't burn Whisper calls
+   *  (or, worse, leak the bot via "Whisper not enabled" replies) on messages
+   *  that would be filtered out anyway.
+   *
+   *  Audio carries no `@bloby` text trigger, so in assistant mode we only
+   *  transcribe when the audio is admin's self-chat command. */
+  private shouldProcessWhatsAppAudio(fromMe: boolean, isSelfChat: boolean, isGroup: boolean): boolean {
+    const channelConfig = this.getChannelConfig('whatsapp');
+    if (!channelConfig) return false;
+
+    const mode = channelConfig.mode || 'channel';
+
+    // Group gating mirrors handleInboundMessage.
+    if (isGroup) {
+      if (mode === 'channel') return false;
+      if (!channelConfig.allowGroups) return false;
+    }
+
+    if (mode === 'channel') return fromMe && isSelfChat;
+    if (mode === 'assistant') return fromMe && isSelfChat;
+    if (mode === 'business') {
+      // Outbound non-self-chat messages are filtered out — same as handleInboundMessage.
+      if (fromMe && !isSelfChat) return false;
+      return true;
+    }
+    return false;
+  }
+
   /** Handle an incoming message from any channel — debounces rapid messages from the same sender.
    *
    * Per-mode behavior is decided here. To add a new mode: extend the gating block below

package/supervisor/channels/whatsapp.ts

@@ -48,6 +48,16 @@ export type OnWhatsAppMessage = (
 /** Callback to transcribe audio via whisper */
 export type TranscribeFn = (audioBase64: string) => Promise<string | null>;
 
+/** Callback that decides whether an audio message warrants transcription.
+ * Returning false makes the channel silently skip the audio (no Whisper call,
+ * no "Whisper not enabled" reply) — used to avoid leaking the bot in modes
+ * where the message would be filtered out downstream anyway. */
+export type ShouldTranscribeAudioFn = (
+  fromMe: boolean,
+  isSelfChat: boolean,
+  isGroup: boolean,
+) => boolean;
+
 export class WhatsAppChannel implements ChannelProvider {
   readonly type: ChannelType = 'whatsapp';
 
@@ -58,6 +68,7 @@ export class WhatsAppChannel implements ChannelProvider {
   private onMessage: OnWhatsAppMessage;
   private onStatusChange: (status: ChannelStatus) => void;
   private transcribe: TranscribeFn | null = null;
+  private shouldTranscribeAudio: ShouldTranscribeAudioFn | null = null;
   private reconnectTimer: ReturnType<typeof setTimeout> | null = null;
   private intentionalDisconnect = false;
 
@@ -76,10 +87,12 @@ export class WhatsAppChannel implements ChannelProvider {
     onMessage: OnWhatsAppMessage,
     onStatusChange: (status: ChannelStatus) => void,
     transcribe?: TranscribeFn,
+    shouldTranscribeAudio?: ShouldTranscribeAudioFn,
   ) {
     this.onMessage = onMessage;
     this.onStatusChange = onStatusChange;
     this.transcribe = transcribe || null;
+    this.shouldTranscribeAudio = shouldTranscribeAudio || null;
   }
 
   async connect(): Promise<void> {
@@ -441,6 +454,29 @@ export class WhatsAppChannel implements ChannelProvider {
           continue;
         }
 
+        // Resolve sender/chat identity up front so audio gating can consult mode/role.
+        const fromMe = msg.key.fromMe || false;
+        const rawSender = msg.key.remoteJid || '';
+        const participant = msg.key.participant || '';
+        const isGroup = rawSender.endsWith('@g.us');
+
+        // chatJid: where to reply (group JID for groups, peer JID otherwise).
+        const chatJid = rawSender;
+
+        // The actual sender JID:
+        //   - groups: always `participant` (remoteJid is the group)
+        //   - 1:1:   `participant` if Baileys provided one (newer protocol), else remoteJid
+        const actualSender = isGroup
+          ? participant || rawSender
+          : (participant || rawSender);
+
+        // Translate LID JIDs to phone JIDs (only handles our own LID)
+        const sender = this.translateJid(actualSender);
+        const pushName = msg.pushName || undefined;
+
+        // Self-chat: only meaningful for 1:1 — remoteJid is our own number AND no participant.
+        const isSelfChat = !isGroup && !participant && this.ownPhoneJid !== null && this.translateJid(rawSender) === this.ownPhoneJid;
+
         // Extract text — or transcribe audio if it's a voice note
         let rawText = this.extractText(msg.message);
         const images: WhatsAppImageAttachment[] = [];
@@ -459,6 +495,13 @@ export class WhatsAppChannel implements ChannelProvider {
         }
 
         if (!rawText && this.isAudioMessage(msg.message)) {
+          // Mode-aware gate: don't transcribe (and don't reveal the bot with a
+          // "Whisper not enabled" reply) when the message would be filtered out
+          // downstream — e.g. a friend's voice note in assistant mode.
+          if (this.shouldTranscribeAudio && !this.shouldTranscribeAudio(fromMe, isSelfChat, isGroup)) {
+            log.info(`[whatsapp] Audio skipped by mode gate (fromMe=${fromMe}, selfChat=${isSelfChat}, group=${isGroup})`);
+            continue;
+          }
           // Voice note / audio — download and transcribe
           if (!this.transcribe) {
             log.info('[whatsapp] Audio message received but no transcribe function configured — skipping');
@@ -494,28 +537,6 @@ export class WhatsAppChannel implements ChannelProvider {
         // Escape special characters to prevent prompt injection via message content
         const text = this.escapeMessageText(rawText);
 
-        const fromMe = msg.key.fromMe || false;
-        const rawSender = msg.key.remoteJid || '';
-        const participant = msg.key.participant || '';
-        const isGroup = rawSender.endsWith('@g.us');
-
-        // chatJid: where to reply (group JID for groups, peer JID otherwise).
-        const chatJid = rawSender;
-
-        // The actual sender JID:
-        //   - groups: always `participant` (remoteJid is the group)
-        //   - 1:1:   `participant` if Baileys provided one (newer protocol), else remoteJid
-        const actualSender = isGroup
-          ? participant || rawSender
-          : (participant || rawSender);
-
-        // Translate LID JIDs to phone JIDs (only handles our own LID)
-        const sender = this.translateJid(actualSender);
-        const pushName = msg.pushName || undefined;
-
-        // Self-chat: only meaningful for 1:1 — remoteJid is our own number AND no participant.
-        const isSelfChat = !isGroup && !participant && this.ownPhoneJid !== null && this.translateJid(rawSender) === this.ownPhoneJid;
-
         log.info(`[whatsapp] Message from ${sender} (chat=${chatJid}, group=${isGroup}, fromMe=${fromMe}, selfChat=${isSelfChat}, images=${images.length}): ${text.slice(0, 80)}`);
 
         this.onMessage(sender, pushName, text, fromMe, isSelfChat, chatJid, isGroup, images.length > 0 ? images : undefined);