feat: improve Telegram conversation heuristics (#14)

Co-authored-by: openhands <openhands@all-hands.dev>

tests/test_bot.py

@@ -44,6 +44,7 @@ def _make_update(
     message = MagicMock(spec=Message)
     message.text = text
     message.reply_text = AsyncMock()
+    message.reply_poll = AsyncMock()
     message.message_thread_id = thread_id
 
     chat = MagicMock(spec=Chat)
@@ -185,6 +186,62 @@ async def test_message_handler_calls_llm_and_replies():
     assert _history[77][1]["role"] == "assistant"
 
 
+@pytest.mark.asyncio
+async def test_message_handler_sends_multiple_reply_messages():
+    settings = _make_settings()
+    mock_llm = MagicMock(spec=LLMClient)
+    mock_llm.chat = AsyncMock(
+        return_value="[MESSAGE]\nFirst thought.\n[MESSAGE]\nWhat outcome do you want?"
+    )
+
+    update = _make_update(user_id=78, text="Can you help with this vague thing?")
+    ctx = _make_context(settings, llm=mock_llm)
+
+    _history[78].clear()
+    await message_handler(update, ctx)
+
+    assert update.message.reply_text.await_count == 2
+    assert update.message.reply_text.await_args_list[0].args[0] == "First thought."
+    assert update.message.reply_text.await_args_list[1].args[0] == "What outcome do you want?"
+    update.message.reply_poll.assert_not_awaited()
+    assert _history[78][1]["content"] == "First thought.\n\nWhat outcome do you want?"
+
+
+@pytest.mark.asyncio
+async def test_message_handler_sends_native_poll_from_directive():
+    settings = _make_settings()
+    mock_llm = MagicMock(spec=LLMClient)
+    mock_llm.chat = AsyncMock(
+        return_value=(
+            "[MESSAGE]\nI can turn that into a vote.\n"
+            "[POLL]\n"
+            "question: Which implementation should we choose?\n"
+            "- Minimal change\n"
+            "- Full redesign\n"
+            "[/POLL]\n"
+            "[MESSAGE]\nI'll use the winning option."
+        )
+    )
+
+    update = _make_update(user_id=79, text="Should we do a minimal change or redesign?")
+    ctx = _make_context(settings, llm=mock_llm)
+
+    _history[79].clear()
+    await message_handler(update, ctx)
+
+    assert update.message.reply_text.await_count == 2
+    update.message.reply_poll.assert_awaited_once_with(
+        question="Which implementation should we choose?",
+        options=["Minimal change", "Full redesign"],
+        is_anonymous=False,
+    )
+    assert _history[79][1]["content"] == (
+        "I can turn that into a vote.\n\n"
+        "Poll: Which implementation should we choose? (Minimal change; Full redesign)\n\n"
+        "I'll use the winning option."
+    )
+
+
 @pytest.mark.asyncio
 async def test_message_handler_non_thread_trims_history():
     """Non-thread history is trimmed to _MAX_HISTORY turns."""