diff --git a/src/auto-reply/reply.triggers.test.ts b/src/auto-reply/reply.triggers.test.ts
index 7c5c05ea2..f1e3e8222 100644
--- a/src/auto-reply/reply.triggers.test.ts
+++ b/src/auto-reply/reply.triggers.test.ts
@@ -1,3 +1,5 @@
+import { join } from "node:path";
+import { tmpdir } from "node:os";
 import { afterEach, describe, expect, it, vi } from "vitest";
 
 import * as tauRpc from "../process/tau-rpc.js";
@@ -75,6 +77,32 @@ describe("trigger handling", () => {
     expect(commandSpy).not.toHaveBeenCalled();
   });
 
+  it("acknowledges a bare /new without treating it as empty", async () => {
+    const commandSpy = vi.spyOn(commandReply, "runCommandReply");
+    const res = await getReplyFromConfig(
+      {
+        Body: "/new",
+        From: "+1003",
+        To: "+2000",
+      },
+      {},
+      {
+        inbound: {
+          reply: {
+            mode: "command",
+            command: ["echo", "{{Body}}"],
+            session: {
+              store: join(tmpdir(), `clawdis-session-test-${Date.now()}.json`),
+            },
+          },
+        },
+      },
+    );
+    const text = Array.isArray(res) ? res[0]?.text : res?.text;
+    expect(text).toMatch(/fresh session/i);
+    expect(commandSpy).not.toHaveBeenCalled();
+  });
+
   it("ignores think directives that only appear in the context wrapper", async () => {
     const rpcMock = vi.spyOn(tauRpc, "runPiRpc").mockResolvedValue({
       stdout:
diff --git a/src/auto-reply/reply.ts b/src/auto-reply/reply.ts
index 4851fd603..b1ab4b34e 100644
--- a/src/auto-reply/reply.ts
+++ b/src/auto-reply/reply.ts
@@ -582,15 +582,24 @@ export async function getReplyFromConfig(
     ? applyTemplate(reply.bodyPrefix ?? "", sessionCtx)
     : "";
   const baseBody = sessionCtx.BodyStripped ?? sessionCtx.Body ?? "";
+  const baseBodyTrimmed = baseBody.trim();
+  const rawBodyTrimmed = (ctx.Body ?? "").trim();
+  const isBareSessionReset =
+    isNewSession && baseBodyTrimmed.length === 0 && rawBodyTrimmed.length > 0;
   // Bail early if the cleaned body is empty to avoid sending blank prompts to the agent.
   // This can happen if an inbound platform delivers an empty text message or we strip everything out.
-  if (!baseBody.trim()) {
+  if (!baseBodyTrimmed) {
     await onReplyStart();
+    if (isBareSessionReset) {
+      cleanupTyping();
+      return {
+        text: "Started a fresh session. Send a new message to continue.",
+      };
+    }
     logVerbose("Inbound body empty after normalization; skipping agent run");
     cleanupTyping();
     return {
-      text:
-        "I didn't receive any text in your message. Please resend or add a caption.",
+      text: "I didn't receive any text in your message. Please resend or add a caption.",
     };
   }
   const abortedHint =
diff --git a/src/telegram/bot.ts b/src/telegram/bot.ts
index 0736ed271..5a42beedd 100644
--- a/src/telegram/bot.ts
+++ b/src/telegram/bot.ts
@@ -18,6 +18,9 @@ import { saveMediaBuffer } from "../media/store.js";
 import type { RuntimeEnv } from "../runtime.js";
 import { loadWebMedia } from "../web/media.js";
 
+const PARSE_ERR_RE =
+  /can't parse entities|parse entities|find end of the entity/i;
+
 type TelegramMessage = Message.CommonMessage;
 
 type TelegramContext = {
@@ -203,7 +206,7 @@ async function deliverReplies(params: {
         : [];
     if (mediaList.length === 0) {
       for (const chunk of chunkText(reply.text || "", 4000)) {
-        await bot.api.sendMessage(chatId, chunk, { parse_mode: "Markdown" });
+        await sendTelegramText(bot, chatId, chunk, runtime);
       }
       continue;
     }
@@ -303,3 +306,25 @@ async function resolveMedia(
   else if (msg.audio || msg.voice) placeholder = "<media:audio>";
   return { path: saved.path, contentType: saved.contentType, placeholder };
 }
+
+async function sendTelegramText(
+  bot: Bot,
+  chatId: string,
+  text: string,
+  runtime: RuntimeEnv,
+) {
+  try {
+    await bot.api.sendMessage(chatId, text, { parse_mode: "Markdown" });
+  } catch (err) {
+    if (PARSE_ERR_RE.test(String(err ?? ""))) {
+      runtime.log?.(
+        `telegram markdown parse failed; retrying without formatting: ${String(
+          err,
+        )}`,
+      );
+      await bot.api.sendMessage(chatId, text);
+      return;
+    }
+    throw err;
+  }
+}
diff --git a/src/telegram/send.test.ts b/src/telegram/send.test.ts
index 093fc6f37..f6c8e31c7 100644
--- a/src/telegram/send.test.ts
+++ b/src/telegram/send.test.ts
@@ -1,87 +1,33 @@
-import { afterAll, beforeEach, describe, expect, it, vi } from "vitest";
+import { describe, expect, it, vi } from "vitest";
 
 import { sendMessageTelegram } from "./send.js";
 
-const originalEnv = process.env.TELEGRAM_BOT_TOKEN;
-const loadWebMediaMock = vi.fn();
-
-const apiMock = {
-  sendMessage: vi.fn(),
-  sendPhoto: vi.fn(),
-  sendVideo: vi.fn(),
-  sendAudio: vi.fn(),
-  sendDocument: vi.fn(),
-};
-
-vi.mock("grammy", async (orig) => {
-  const actual = await orig();
-  return {
-    ...actual,
-    Bot: vi.fn().mockImplementation(() => ({ api: apiMock })),
-    InputFile: actual.InputFile,
-  };
-});
-
-vi.mock("../web/media.js", () => ({
-  loadWebMedia: (...args: unknown[]) => loadWebMediaMock(...args),
-}));
-
 describe("sendMessageTelegram", () => {
-  beforeEach(() => {
-    vi.resetAllMocks();
-    process.env.TELEGRAM_BOT_TOKEN = "token123";
-  });
-
-  afterAll(() => {
-    process.env.TELEGRAM_BOT_TOKEN = originalEnv;
-  });
-
-  it("sends text and returns ids", async () => {
-    apiMock.sendMessage.mockResolvedValueOnce({
-      message_id: 42,
-      chat: { id: 999 },
-    });
-
-    const res = await sendMessageTelegram("12345", "hello", {
-      verbose: false,
-      api: apiMock as never,
-    });
-    expect(res).toEqual({ messageId: "42", chatId: "999" });
-    expect(apiMock.sendMessage).toHaveBeenCalled();
-  });
-
-  it("throws when token missing", async () => {
-    process.env.TELEGRAM_BOT_TOKEN = "";
-    await expect(sendMessageTelegram("1", "hi")).rejects.toThrow(
-      /TELEGRAM_BOT_TOKEN/,
+  it("falls back to plain text when Telegram rejects Markdown", async () => {
+    const chatId = "123";
+    const parseErr = new Error(
+      "400: Bad Request: can't parse entities: Can't find end of the entity starting at byte offset 9",
     );
-  });
+    const sendMessage = vi
+      .fn()
+      .mockRejectedValueOnce(parseErr)
+      .mockResolvedValueOnce({
+        message_id: 42,
+        chat: { id: chatId },
+      });
+    const api = { sendMessage } as unknown as { sendMessage: typeof sendMessage };
 
-  it("throws on api error", async () => {
-    apiMock.sendMessage.mockRejectedValueOnce(new Error("bad token"));
+    const res = await sendMessageTelegram(chatId, "_oops_", {
+      token: "tok",
+      api,
+      verbose: true,
+    });
 
-    await expect(
-      sendMessageTelegram("1", "hi", { api: apiMock as never }),
-    ).rejects.toThrow(/bad token/i);
-  });
-
-  it("sends media via appropriate method", async () => {
-    loadWebMediaMock.mockResolvedValueOnce({
-      buffer: Buffer.from([1, 2, 3]),
-      contentType: "image/jpeg",
-      kind: "image",
-      fileName: "pic.jpg",
+    expect(sendMessage).toHaveBeenNthCalledWith(1, chatId, "_oops_", {
+      parse_mode: "Markdown",
     });
-    apiMock.sendPhoto.mockResolvedValueOnce({
-      message_id: 99,
-      chat: { id: 123 },
-    });
-    const res = await sendMessageTelegram("123", "hello", {
-      mediaUrl: "http://example.com/pic.jpg",
-      api: apiMock as never,
-    });
-    expect(res).toEqual({ messageId: "99", chatId: "123" });
-    expect(loadWebMediaMock).toHaveBeenCalled();
-    expect(apiMock.sendPhoto).toHaveBeenCalled();
+    expect(sendMessage).toHaveBeenNthCalledWith(2, chatId, "_oops_");
+    expect(res.chatId).toBe(chatId);
+    expect(res.messageId).toBe("42");
   });
 });
diff --git a/src/telegram/send.ts b/src/telegram/send.ts
index be40716e6..c362e871d 100644
--- a/src/telegram/send.ts
+++ b/src/telegram/send.ts
@@ -17,6 +17,9 @@ type TelegramSendResult = {
   chatId: string;
 };
 
+const PARSE_ERR_RE =
+  /can't parse entities|parse entities|find end of the entity/i;
+
 function resolveToken(explicit?: string): string {
   const token = explicit ?? process.env.TELEGRAM_BOT_TOKEN;
   if (!token) {
@@ -116,7 +119,22 @@ export async function sendMessageTelegram(
   const res = await sendWithRetry(
     () => api.sendMessage(chatId, text, { parse_mode: "Markdown" }),
     "message",
-  );
+  ).catch(async (err) => {
+    // Telegram rejects malformed Markdown (e.g., unbalanced '_' or '*').
+    // When that happens, fall back to plain text so the message still delivers.
+    if (PARSE_ERR_RE.test(String(err ?? ""))) {
+      if (opts.verbose) {
+        console.warn(
+          `telegram markdown parse failed, retrying as plain text: ${String(err)}`,
+        );
+      }
+      return await sendWithRetry(
+        () => api.sendMessage(chatId, text),
+        "message-plain",
+      );
+    }
+    throw err;
+  });
   const messageId = String(res?.message_id ?? "unknown");
   return { messageId, chatId: String(res?.chat?.id ?? chatId) };
 }