fix(telegram): restore 30-char preview debounce

src/auto-reply/reply/agent-runner-execution.ts

@@ -138,10 +138,7 @@ export async function runAgentTurnWithFallback(params: {
           }
           text = stripped.text;
         }
-        if (
-          isSilentReplyText(text, SILENT_REPLY_TOKEN) ||
-          isSilentReplyPrefixText(text, SILENT_REPLY_TOKEN)
-        ) {
+        if (isSilentReplyText(text, SILENT_REPLY_TOKEN)) {
           return { skip: true };
         }
         if (!text) {
@@ -160,6 +157,9 @@ export async function runAgentTurnWithFallback(params: {
         return { text: sanitized, skip: false };
       };
       const handlePartialForTyping = async (payload: ReplyPayload): Promise<string | undefined> => {
+        if (isSilentReplyPrefixText(payload.text, SILENT_REPLY_TOKEN)) {
+          return undefined;
+        }
         const { text, skip } = normalizeStreamingText(payload);
         if (skip || !text) {
           return undefined;

src/auto-reply/reply/agent-runner.runreplyagent.test.ts

@@ -403,6 +403,30 @@ describe("runReplyAgent typing (heartbeat)", () => {
     expect(typing.startTypingLoop).not.toHaveBeenCalled();
   });
 
+  it("does not suppress partial streaming for normal 'No' prefixes", async () => {
+    const onPartialReply = vi.fn();
+    state.runEmbeddedPiAgentMock.mockImplementationOnce(async (params: AgentRunParams) => {
+      await params.onPartialReply?.({ text: "No" });
+      await params.onPartialReply?.({ text: "No, that is valid" });
+      return { payloads: [{ text: "No, that is valid" }], meta: {} };
+    });
+
+    const { run, typing } = createMinimalRun({
+      opts: { isHeartbeat: false, onPartialReply },
+      typingMode: "message",
+    });
+    await run();
+
+    expect(onPartialReply).toHaveBeenCalledTimes(2);
+    expect(onPartialReply).toHaveBeenNthCalledWith(1, { text: "No", mediaUrls: undefined });
+    expect(onPartialReply).toHaveBeenNthCalledWith(2, {
+      text: "No, that is valid",
+      mediaUrls: undefined,
+    });
+    expect(typing.startTypingOnText).toHaveBeenCalled();
+    expect(typing.startTypingLoop).not.toHaveBeenCalled();
+  });
+
   it("does not start typing on assistant message start without prior text in message mode", async () => {
     state.runEmbeddedPiAgentMock.mockImplementationOnce(async (params: AgentRunParams) => {
       await params.onAssistantMessageStart?.();

src/auto-reply/tokens.ts

@@ -30,5 +30,11 @@ export function isSilentReplyPrefixText(
   if (!normalized) {
     return false;
   }
+  if (!normalized.includes("_")) {
+    return false;
+  }
+  if (/[^A-Z_]/.test(normalized)) {
+    return false;
+  }
   return token.toUpperCase().startsWith(normalized);
 }

src/telegram/bot-message-dispatch.test.ts

@@ -191,7 +191,7 @@ describe("dispatchTelegramMessage draft streaming", () => {
       expect.objectContaining({
         chatId: 123,
         thread: { id: 777, scope: "dm" },
-        minInitialChars: 1,
+        minInitialChars: 30,
       }),
     );
     expect(draftStream.update).toHaveBeenCalledWith("Hello");
@@ -212,7 +212,7 @@ describe("dispatchTelegramMessage draft streaming", () => {
     expect(draftStream.clear).toHaveBeenCalledTimes(1);
   });
 
-  it("uses immediate preview updates for legacy block stream mode", async () => {
+  it("uses 30-char preview debounce for legacy block stream mode", async () => {
     const draftStream = createDraftStream();
     createTelegramDraftStream.mockReturnValue(draftStream);
     dispatchReplyWithBufferedBlockDispatcher.mockImplementation(
@@ -228,7 +228,7 @@ describe("dispatchTelegramMessage draft streaming", () => {
 
     expect(createTelegramDraftStream).toHaveBeenCalledWith(
       expect.objectContaining({
-        minInitialChars: 1,
+        minInitialChars: 30,
       }),
     );
   });

src/telegram/bot-message-dispatch.ts

@@ -147,8 +147,7 @@ export const dispatchTelegramMessage = async ({
   const canStreamReasoningDraft = canStreamAnswerDraft || streamReasoningDraft;
   const draftReplyToMessageId =
     replyToMode !== "off" && typeof msg.message_id === "number" ? msg.message_id : undefined;
-  const draftMinInitialChars =
-    previewStreamingEnabled || streamReasoningDraft ? 1 : DRAFT_MIN_INITIAL_CHARS;
+  const draftMinInitialChars = DRAFT_MIN_INITIAL_CHARS;
   const mediaLocalRoots = getAgentScopedMediaLocalRoots(cfg, route.agentId);
   type LaneName = "answer" | "reasoning";
   type DraftLaneState = {