feat(telegram-remote): show transcript in Telegram only, not to AI

packages/telegram-remote/README.md

@@ -9,7 +9,7 @@ Turn any Telegram chat into a pocket terminal for your harness: chat with your A
 ## ✨ Features
 
 - 💬 **Chat from Telegram** — send a message, get a live-streamed structured reply (thinking, tools, final answer) edited in place
-- 🎙️ **Voice & audio** — send a voice message or audio file; it is transcribed through your home-lab Whisper server (default `large-v2` on CUDA GPU) and the text is sent to your AI
+- 🎙️ **Voice & audio** — send a voice message or audio file; it is transcribed through your home-lab Whisper server (default `large-v2` on CUDA GPU) and the text is shown in Telegram (not sent to the AI agent)
 - 🎛 **Full harness control** — 49 commands: chats, models, sessions, subagents, goals, jobs, files, PowerShell, exports, presets, skills, settings, credentials, permissions, and more
 - 📡 **Live state** — status, running turns, queued messages (steer / edit / remove), background jobs
 - 🔐 **Permission-aware** — sandbox read/write/full control per chat, approval buttons for risky tools

packages/telegram-remote/lib/index.js

@@ -505,8 +505,8 @@ function pickAudioFilename(audio, mimeType) {
 
 /**
  * Transcribe a voice message / audio file / audio document via the home-lab
- * Whishper server, show the transcript, and queue it into the chat session
+ * Whishper server and show the transcript in Telegram. The transcript is
- * like a normal user message.
+ * deliberately NOT sent to the AI agent as a prompt.
  */
 async function handleAudio(state, chatId, userId, audio, messageId) {
   let statusMessageId = null;
@@ -550,27 +550,7 @@ async function handleAudio(state, chatId, userId, audio, messageId) {
       state.log("audio transcript edit failed: " + editError.message);
     }
 
-    const cs = state.chatState(chatId);
+    state.log("audio transcribed (" + state.config.whisperModel + ") for chat " + chatId + " by " + userId);
-    if (messageId != null) cs.lastUserMessageId = messageId;
-    if (!cs.sessionId) {
-      try {
-        const created = await callApi(state.ctx, "sessions", "create", { cwd: state.config.workspaceRoot || process.cwd() });
-        cs.sessionId = created.sessionId;
-        state.saveState();
-      } catch {}
-    }
-    if (!cs.sessionId) {
-      await state.bot.send(chatId, "No chat is open yet — tap 💬 New chat first.");
-      return;
-    }
-    await callApi(state.ctx, "sessions", "prompt", {
-      sessionId: cs.sessionId,
-      mode: "queue",
-      content: [{ type: "text", text: trimmed }],
-      clientTimeZone: (() => { try { return Intl.DateTimeFormat().resolvedOptions().timeZone; } catch { return "UTC"; } })(),
-    });
-    state.notePrompt(cs.sessionId, trimmed);
-    state.log("audio transcribed (" + state.config.whisperModel + ") and sent to " + cs.sessionId + " by " + userId);
   } catch (error) {
     state.log("audio handling failed: " + (error.stack ?? error.message));
     const message = "⚠️ Couldn't transcribe the audio: " + esc(error.message);