fix(telegram): also check caption for bot media replies

Address Greptile review feedback: bot media messages (photo/video) use
caption instead of text, so they would be incorrectly classified as
system messages.  Add !caption guard to the system message check.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/telegram/bot-message-context.implicit-mention.test.ts

@@ -10,6 +10,7 @@ describe("buildTelegramMessageContext implicitMention forum system messages", ()
    */
   async function buildGroupReplyCtx(params: {
     replyToMessageText?: string;
+    replyToMessageCaption?: string;
     replyFromIsBot?: boolean;
     replyFromId?: number;
   }) {
@@ -24,6 +25,9 @@ describe("buildTelegramMessageContext implicitMention forum system messages", ()
         reply_to_message: {
           message_id: 1,
           text: params.replyToMessageText ?? undefined,
+          ...(params.replyToMessageCaption != null
+            ? { caption: params.replyToMessageCaption }
+            : {}),
           from: {
             id: params.replyFromId ?? BOT_ID,
             first_name: "OpenClaw",
@@ -85,6 +89,19 @@ describe("buildTelegramMessageContext implicitMention forum system messages", ()
     expect(ctx?.ctxPayload?.WasMentioned).toBe(true);
   });
 
+  it("DOES trigger implicitMention for bot media messages with caption (not a system message)", async () => {
+    // Media messages from the bot have caption but no text — they should
+    // still count as real bot replies, not system messages.
+    const ctx = await buildGroupReplyCtx({
+      replyToMessageText: undefined,
+      replyToMessageCaption: "Check out this image",
+      replyFromIsBot: true,
+    });
+
+    expect(ctx).not.toBeNull();
+    expect(ctx?.ctxPayload?.WasMentioned).toBe(true);
+  });
+
   it("does NOT trigger implicitMention when reply is from a different user", async () => {
     const ctx = await buildGroupReplyCtx({
       replyToMessageText: "some message",