From c81d952f9f73aad9127c6f12f93d84d48687e741 Mon Sep 17 00:00:00 2001
From: Cyber MacGeddon <cybermaggedon@gmail.com>
Date: Wed, 26 Nov 2025 19:33:40 +0000
Subject: [PATCH] Prompt tests + LLM tests

---
 .../test_prompt_streaming_integration.py      | 389 ++++++++++++++++++
 ...t_text_completion_streaming_integration.py | 365 ++++++++++++++++
 tests/test_parser_debug.py                    |  51 +++
 3 files changed, 805 insertions(+)
 create mode 100644 tests/integration/test_prompt_streaming_integration.py
 create mode 100644 tests/integration/test_text_completion_streaming_integration.py
 create mode 100644 tests/test_parser_debug.py

diff --git a/tests/integration/test_prompt_streaming_integration.py b/tests/integration/test_prompt_streaming_integration.py
new file mode 100644
index 00000000..65b3d68a
--- /dev/null
+++ b/tests/integration/test_prompt_streaming_integration.py
@@ -0,0 +1,389 @@
+"""
+Integration tests for Prompt Service Streaming Functionality
+
+These tests verify the streaming behavior of the Prompt service,
+testing how it coordinates between templates and text completion streaming.
+"""
+
+import pytest
+from unittest.mock import AsyncMock, MagicMock
+from trustgraph.prompt.template.service import Processor
+from trustgraph.schema import PromptRequest, PromptResponse, TextCompletionResponse
+from tests.utils.streaming_assertions import (
+    assert_streaming_chunks_valid,
+    assert_callback_invoked,
+)
+
+
+@pytest.mark.integration
+class TestPromptStreaming:
+    """Integration tests for Prompt service streaming"""
+
+    @pytest.fixture
+    def mock_flow_context_streaming(self):
+        """Mock flow context with streaming text completion support"""
+        context = MagicMock()
+
+        # Mock text completion client with streaming
+        text_completion_client = AsyncMock()
+
+        async def streaming_request(request, recipient=None, timeout=600):
+            """Simulate streaming text completion"""
+            if request.streaming and recipient:
+                # Simulate streaming chunks
+                chunks = [
+                    "Machine", " learning", " is", " a", " field",
+                    " of", " artificial", " intelligence", "."
+                ]
+
+                for i, chunk_text in enumerate(chunks):
+                    is_final = (i == len(chunks) - 1)
+                    response = TextCompletionResponse(
+                        response=chunk_text,
+                        error=None,
+                        end_of_stream=is_final
+                    )
+                    final = await recipient(response)
+                    if final:
+                        break
+
+                # Final empty chunk
+                await recipient(TextCompletionResponse(
+                    response="",
+                    error=None,
+                    end_of_stream=True
+                ))
+
+        text_completion_client.request = streaming_request
+
+        # Mock response producer
+        response_producer = AsyncMock()
+
+        def context_router(service_name):
+            if service_name == "text-completion-request":
+                return text_completion_client
+            elif service_name == "response":
+                return response_producer
+            else:
+                return AsyncMock()
+
+        context.side_effect = context_router
+        return context
+
+    @pytest.fixture
+    def mock_prompt_manager(self):
+        """Mock PromptManager with simple template"""
+        manager = MagicMock()
+
+        async def invoke_template(kind, input_vars, llm_function):
+            """Simulate template invocation"""
+            # Call the LLM function with simple prompts
+            system = "You are a helpful assistant."
+            prompt = f"Question: {input_vars.get('question', 'test')}"
+            result = await llm_function(system, prompt)
+            return result
+
+        manager.invoke = invoke_template
+        return manager
+
+    @pytest.fixture
+    def prompt_processor_streaming(self, mock_prompt_manager):
+        """Create Prompt processor with streaming support"""
+        processor = MagicMock()
+        processor.manager = mock_prompt_manager
+        processor.config_key = "prompt"
+
+        # Bind the actual on_request method
+        processor.on_request = Processor.on_request.__get__(processor, Processor)
+
+        return processor
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_basic(self, prompt_processor_streaming, mock_flow_context_streaming):
+        """Test basic prompt streaming functionality"""
+        # Arrange
+        request = PromptRequest(
+            id="kg_prompt",
+            terms={"question": '"What is machine learning?"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-123"}
+
+        consumer = MagicMock()
+
+        # Act
+        await prompt_processor_streaming.on_request(
+            message, consumer, mock_flow_context_streaming
+        )
+
+        # Assert
+        # Verify response producer was called multiple times (for streaming chunks)
+        response_producer = mock_flow_context_streaming("response")
+        assert response_producer.send.call_count > 0
+
+        # Verify streaming chunks were sent
+        calls = response_producer.send.call_args_list
+        assert len(calls) > 1  # Should have multiple chunks
+
+        # Check that responses have end_of_stream flag
+        for call in calls:
+            response = call.args[0]
+            assert isinstance(response, PromptResponse)
+            assert hasattr(response, 'end_of_stream')
+
+        # Last response should have end_of_stream=True
+        last_call = calls[-1]
+        last_response = last_call.args[0]
+        assert last_response.end_of_stream is True
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_non_streaming_mode(self, prompt_processor_streaming,
+                                                         mock_flow_context_streaming):
+        """Test prompt service in non-streaming mode"""
+        # Arrange
+        request = PromptRequest(
+            id="kg_prompt",
+            terms={"question": '"What is AI?"'},
+            streaming=False  # Non-streaming
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-456"}
+
+        consumer = MagicMock()
+
+        # Mock non-streaming text completion
+        text_completion_client = mock_flow_context_streaming("text-completion-request")
+
+        async def non_streaming_text_completion(system, prompt, streaming=False):
+            return "AI is the simulation of human intelligence in machines."
+
+        text_completion_client.text_completion = non_streaming_text_completion
+
+        # Act
+        await prompt_processor_streaming.on_request(
+            message, consumer, mock_flow_context_streaming
+        )
+
+        # Assert
+        # Verify response producer was called once (non-streaming)
+        response_producer = mock_flow_context_streaming("response")
+        # Note: In non-streaming mode, the service sends a single response
+        assert response_producer.send.call_count >= 1
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_chunk_forwarding(self, prompt_processor_streaming,
+                                                       mock_flow_context_streaming):
+        """Test that prompt service forwards chunks immediately"""
+        # Arrange
+        request = PromptRequest(
+            id="test_prompt",
+            terms={"question": '"Test query"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-789"}
+
+        consumer = MagicMock()
+
+        # Act
+        await prompt_processor_streaming.on_request(
+            message, consumer, mock_flow_context_streaming
+        )
+
+        # Assert
+        # Verify chunks were forwarded with proper structure
+        response_producer = mock_flow_context_streaming("response")
+        calls = response_producer.send.call_args_list
+
+        for call in calls:
+            response = call.args[0]
+            # Each response should have text and end_of_stream fields
+            assert hasattr(response, 'text')
+            assert hasattr(response, 'end_of_stream')
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_error_handling(self, prompt_processor_streaming):
+        """Test error handling during streaming"""
+        # Arrange
+        context = MagicMock()
+
+        # Mock text completion client that raises an error
+        text_completion_client = AsyncMock()
+
+        async def failing_request(request, recipient=None, timeout=600):
+            if recipient:
+                # Send error response
+                error_response = TextCompletionResponse(
+                    response="",
+                    error=MagicMock(message="Text completion error"),
+                    end_of_stream=True
+                )
+                await recipient(error_response)
+
+        text_completion_client.request = failing_request
+
+        def context_router(service_name):
+            if service_name == "text-completion-request":
+                return text_completion_client
+            elif service_name == "response":
+                return AsyncMock()
+            else:
+                return AsyncMock()
+
+        context.side_effect = context_router
+
+        request = PromptRequest(
+            id="test_prompt",
+            terms={"question": '"Test"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-error"}
+
+        consumer = MagicMock()
+
+        # Act & Assert
+        with pytest.raises(RuntimeError) as exc_info:
+            await prompt_processor_streaming.on_request(message, consumer, context)
+
+        assert "Text completion error" in str(exc_info.value)
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_preserves_message_id(self, prompt_processor_streaming,
+                                                           mock_flow_context_streaming):
+        """Test that message IDs are preserved through streaming"""
+        # Arrange
+        message_id = "unique-test-id-12345"
+
+        request = PromptRequest(
+            id="test_prompt",
+            terms={"question": '"Test"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": message_id}
+
+        consumer = MagicMock()
+
+        # Act
+        await prompt_processor_streaming.on_request(
+            message, consumer, mock_flow_context_streaming
+        )
+
+        # Assert
+        # Verify all responses were sent with the correct message ID
+        response_producer = mock_flow_context_streaming("response")
+        calls = response_producer.send.call_args_list
+
+        for call in calls:
+            properties = call.kwargs.get('properties')
+            assert properties is not None
+            assert properties['id'] == message_id
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_empty_response_handling(self, prompt_processor_streaming):
+        """Test handling of empty responses during streaming"""
+        # Arrange
+        context = MagicMock()
+
+        # Mock text completion that sends empty chunks
+        text_completion_client = AsyncMock()
+
+        async def empty_streaming_request(request, recipient=None, timeout=600):
+            if request.streaming and recipient:
+                # Send empty chunk followed by final marker
+                await recipient(TextCompletionResponse(
+                    response="",
+                    error=None,
+                    end_of_stream=False
+                ))
+                await recipient(TextCompletionResponse(
+                    response="",
+                    error=None,
+                    end_of_stream=True
+                ))
+
+        text_completion_client.request = empty_streaming_request
+        response_producer = AsyncMock()
+
+        def context_router(service_name):
+            if service_name == "text-completion-request":
+                return text_completion_client
+            elif service_name == "response":
+                return response_producer
+            else:
+                return AsyncMock()
+
+        context.side_effect = context_router
+
+        request = PromptRequest(
+            id="test_prompt",
+            terms={"question": '"Test"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-empty"}
+
+        consumer = MagicMock()
+
+        # Act
+        await prompt_processor_streaming.on_request(message, consumer, context)
+
+        # Assert
+        # Should still send responses even if empty (including final marker)
+        assert response_producer.send.call_count > 0
+
+        # Last response should have end_of_stream=True
+        last_call = response_producer.send.call_args_list[-1]
+        last_response = last_call.args[0]
+        assert last_response.end_of_stream is True
+
+    @pytest.mark.asyncio
+    async def test_prompt_streaming_concatenation_matches_complete(self, prompt_processor_streaming,
+                                                                     mock_flow_context_streaming):
+        """Test that streaming chunks concatenate to form complete response"""
+        # Arrange
+        request = PromptRequest(
+            id="test_prompt",
+            terms={"question": '"What is ML?"'},
+            streaming=True
+        )
+
+        message = MagicMock()
+        message.value.return_value = request
+        message.properties.return_value = {"id": "test-concat"}
+
+        consumer = MagicMock()
+
+        # Act
+        await prompt_processor_streaming.on_request(
+            message, consumer, mock_flow_context_streaming
+        )
+
+        # Assert
+        # Collect all response texts
+        response_producer = mock_flow_context_streaming("response")
+        calls = response_producer.send.call_args_list
+
+        chunk_texts = []
+        for call in calls:
+            response = call.args[0]
+            if response.text and not response.end_of_stream:
+                chunk_texts.append(response.text)
+
+        # Verify chunks concatenate to expected result
+        full_text = "".join(chunk_texts)
+        assert full_text == "Machine learning is a field of artificial intelligence."
diff --git a/tests/integration/test_text_completion_streaming_integration.py b/tests/integration/test_text_completion_streaming_integration.py
new file mode 100644
index 00000000..7c9ff18a
--- /dev/null
+++ b/tests/integration/test_text_completion_streaming_integration.py
@@ -0,0 +1,365 @@
+"""
+Integration tests for Text Completion Streaming Functionality
+
+These tests verify the streaming behavior of the Text Completion service,
+testing token-by-token response delivery through the complete pipeline.
+"""
+
+import pytest
+from unittest.mock import AsyncMock, MagicMock
+from openai.types.chat import ChatCompletionChunk
+from openai.types.chat.chat_completion_chunk import Choice as StreamChoice, ChoiceDelta
+
+from trustgraph.model.text_completion.openai.llm import Processor
+from trustgraph.base import LlmChunk
+from tests.utils.streaming_assertions import (
+    assert_streaming_chunks_valid,
+    assert_callback_invoked,
+)
+
+
+@pytest.mark.integration
+class TestTextCompletionStreaming:
+    """Integration tests for Text Completion streaming"""
+
+    @pytest.fixture
+    def mock_streaming_openai_client(self, mock_streaming_llm_response):
+        """Mock OpenAI client with streaming support"""
+        client = MagicMock()
+
+        def create_streaming_completion(**kwargs):
+            """Generator that yields streaming chunks"""
+            if kwargs.get('stream', False):
+                # Simulate OpenAI streaming response
+                chunks_text = [
+                    "Machine", " learning", " is", " a", " subset",
+                    " of", " AI", " that", " enables", " computers",
+                    " to", " learn", " from", " data", "."
+                ]
+
+                for text in chunks_text:
+                    delta = ChoiceDelta(content=text, role=None)
+                    choice = StreamChoice(index=0, delta=delta, finish_reason=None)
+                    chunk = ChatCompletionChunk(
+                        id="chatcmpl-streaming",
+                        choices=[choice],
+                        created=1234567890,
+                        model="gpt-3.5-turbo",
+                        object="chat.completion.chunk"
+                    )
+                    yield chunk
+            else:
+                # Non-streaming response (shouldn't happen in these tests)
+                raise ValueError("Expected streaming mode")
+
+        client.chat.completions.create.return_value = create_streaming_completion()
+        return client
+
+    @pytest.fixture
+    def text_completion_processor_streaming(self, mock_streaming_openai_client):
+        """Create text completion processor with streaming support"""
+        processor = MagicMock()
+        processor.default_model = "gpt-3.5-turbo"
+        processor.temperature = 0.7
+        processor.max_output = 1024
+        processor.openai = mock_streaming_openai_client
+
+        # Bind the actual streaming method
+        processor.generate_content_stream = Processor.generate_content_stream.__get__(
+            processor, Processor
+        )
+
+        return processor
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_basic(self, text_completion_processor_streaming,
+                                                     streaming_chunk_collector):
+        """Test basic text completion streaming functionality"""
+        # Arrange
+        system_prompt = "You are a helpful assistant."
+        user_prompt = "What is machine learning?"
+        collector = streaming_chunk_collector()
+
+        # Act - Collect all chunks
+        chunks = []
+        async for chunk in text_completion_processor_streaming.generate_content_stream(
+            system_prompt, user_prompt
+        ):
+            chunks.append(chunk)
+            if chunk.text:
+                await collector.collect(chunk.text)
+
+        # Assert
+        assert len(chunks) > 1  # Should have multiple chunks
+
+        # Verify all chunks are LlmChunk objects
+        for chunk in chunks:
+            assert isinstance(chunk, LlmChunk)
+            assert chunk.model == "gpt-3.5-turbo"
+
+        # Verify last chunk has is_final=True
+        assert chunks[-1].is_final is True
+
+        # Verify we got meaningful content
+        full_text = collector.get_full_text()
+        assert "machine" in full_text.lower() or "learning" in full_text.lower()
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_chunk_structure(self, text_completion_processor_streaming):
+        """Test that streaming chunks have correct structure"""
+        # Arrange
+        system_prompt = "You are a helpful assistant."
+        user_prompt = "Explain AI."
+
+        # Act
+        chunks = []
+        async for chunk in text_completion_processor_streaming.generate_content_stream(
+            system_prompt, user_prompt
+        ):
+            chunks.append(chunk)
+
+        # Assert - Verify chunk structure
+        for i, chunk in enumerate(chunks[:-1]):  # All except last
+            assert isinstance(chunk, LlmChunk)
+            assert chunk.text is not None
+            assert chunk.model == "gpt-3.5-turbo"
+            assert chunk.is_final is False
+
+        # Last chunk should be final marker
+        final_chunk = chunks[-1]
+        assert final_chunk.is_final is True
+        assert final_chunk.model == "gpt-3.5-turbo"
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_concatenation(self, text_completion_processor_streaming):
+        """Test that chunks concatenate to form complete response"""
+        # Arrange
+        system_prompt = "You are a helpful assistant."
+        user_prompt = "What is AI?"
+
+        # Act - Collect all chunk texts
+        chunk_texts = []
+        async for chunk in text_completion_processor_streaming.generate_content_stream(
+            system_prompt, user_prompt
+        ):
+            if chunk.text and not chunk.is_final:
+                chunk_texts.append(chunk.text)
+
+        # Assert
+        full_text = "".join(chunk_texts)
+        assert len(full_text) > 0
+        assert len(chunk_texts) > 1  # Should have multiple chunks
+
+        # Verify completeness - should be a coherent sentence
+        assert full_text == "Machine learning is a subset of AI that enables computers to learn from data."
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_final_marker(self, text_completion_processor_streaming):
+        """Test that final chunk properly marks end of stream"""
+        # Arrange
+        system_prompt = "You are a helpful assistant."
+        user_prompt = "Test query"
+
+        # Act
+        chunks = []
+        async for chunk in text_completion_processor_streaming.generate_content_stream(
+            system_prompt, user_prompt
+        ):
+            chunks.append(chunk)
+
+        # Assert
+        # Should have at least content chunks + final marker
+        assert len(chunks) >= 2
+
+        # Only the last chunk should have is_final=True
+        for chunk in chunks[:-1]:
+            assert chunk.is_final is False
+
+        assert chunks[-1].is_final is True
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_model_parameter(self, mock_streaming_openai_client):
+        """Test that model parameter is preserved in streaming"""
+        # Arrange
+        processor = MagicMock()
+        processor.default_model = "gpt-4"
+        processor.temperature = 0.5
+        processor.max_output = 2048
+        processor.openai = mock_streaming_openai_client
+        processor.generate_content_stream = Processor.generate_content_stream.__get__(
+            processor, Processor
+        )
+
+        # Act
+        chunks = []
+        async for chunk in processor.generate_content_stream("System", "Prompt"):
+            chunks.append(chunk)
+
+        # Assert
+        # Verify OpenAI was called with correct model
+        call_args = mock_streaming_openai_client.chat.completions.create.call_args
+        assert call_args.kwargs['model'] == "gpt-4"
+        assert call_args.kwargs['temperature'] == 0.5
+        assert call_args.kwargs['max_tokens'] == 2048
+        assert call_args.kwargs['stream'] is True
+
+        # Verify chunks have correct model
+        for chunk in chunks:
+            assert chunk.model == "gpt-4"
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_temperature_parameter(self, mock_streaming_openai_client):
+        """Test that temperature parameter is applied in streaming"""
+        # Arrange
+        temperatures = [0.0, 0.5, 1.0, 1.5]
+
+        for temp in temperatures:
+            processor = MagicMock()
+            processor.default_model = "gpt-3.5-turbo"
+            processor.temperature = temp
+            processor.max_output = 1024
+            processor.openai = mock_streaming_openai_client
+            processor.generate_content_stream = Processor.generate_content_stream.__get__(
+                processor, Processor
+            )
+
+            # Act
+            chunks = []
+            async for chunk in processor.generate_content_stream("System", "Prompt"):
+                chunks.append(chunk)
+                if chunk.is_final:
+                    break
+
+            # Assert
+            call_args = mock_streaming_openai_client.chat.completions.create.call_args
+            assert call_args.kwargs['temperature'] == temp
+
+            # Reset mock for next iteration
+            mock_streaming_openai_client.reset_mock()
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_error_propagation(self):
+        """Test that errors during streaming are properly propagated"""
+        # Arrange
+        mock_client = MagicMock()
+
+        def failing_stream(**kwargs):
+            yield from []
+            raise Exception("Streaming error")
+
+        mock_client.chat.completions.create.return_value = failing_stream()
+
+        processor = MagicMock()
+        processor.default_model = "gpt-3.5-turbo"
+        processor.temperature = 0.7
+        processor.max_output = 1024
+        processor.openai = mock_client
+        processor.generate_content_stream = Processor.generate_content_stream.__get__(
+            processor, Processor
+        )
+
+        # Act & Assert
+        with pytest.raises(Exception) as exc_info:
+            async for chunk in processor.generate_content_stream("System", "Prompt"):
+                pass
+
+        assert "Streaming error" in str(exc_info.value)
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_empty_chunks_filtered(self, mock_streaming_openai_client):
+        """Test that empty chunks are handled correctly"""
+        # Arrange - Mock that returns some empty chunks
+        def create_streaming_with_empties(**kwargs):
+            chunks_text = ["Hello", "", " world", "", "!"]
+
+            for text in chunks_text:
+                delta = ChoiceDelta(content=text if text else None, role=None)
+                choice = StreamChoice(index=0, delta=delta, finish_reason=None)
+                chunk = ChatCompletionChunk(
+                    id="chatcmpl-streaming",
+                    choices=[choice],
+                    created=1234567890,
+                    model="gpt-3.5-turbo",
+                    object="chat.completion.chunk"
+                )
+                yield chunk
+
+        mock_streaming_openai_client.chat.completions.create.return_value = create_streaming_with_empties()
+
+        processor = MagicMock()
+        processor.default_model = "gpt-3.5-turbo"
+        processor.temperature = 0.7
+        processor.max_output = 1024
+        processor.openai = mock_streaming_openai_client
+        processor.generate_content_stream = Processor.generate_content_stream.__get__(
+            processor, Processor
+        )
+
+        # Act
+        chunks = []
+        async for chunk in processor.generate_content_stream("System", "Prompt"):
+            chunks.append(chunk)
+
+        # Assert - Only non-empty chunks should be yielded (plus final marker)
+        text_chunks = [c for c in chunks if not c.is_final]
+        assert len(text_chunks) == 3  # "Hello", " world", "!"
+        assert "".join(c.text for c in text_chunks) == "Hello world!"
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_prompt_construction(self, mock_streaming_openai_client):
+        """Test that system and user prompts are correctly combined for streaming"""
+        # Arrange
+        processor = MagicMock()
+        processor.default_model = "gpt-3.5-turbo"
+        processor.temperature = 0.7
+        processor.max_output = 1024
+        processor.openai = mock_streaming_openai_client
+        processor.generate_content_stream = Processor.generate_content_stream.__get__(
+            processor, Processor
+        )
+
+        system_prompt = "You are an expert."
+        user_prompt = "Explain quantum physics."
+
+        # Act
+        chunks = []
+        async for chunk in processor.generate_content_stream(system_prompt, user_prompt):
+            chunks.append(chunk)
+            if chunk.is_final:
+                break
+
+        # Assert - Verify prompts were combined correctly
+        call_args = mock_streaming_openai_client.chat.completions.create.call_args
+        messages = call_args.kwargs['messages']
+        assert len(messages) == 1
+
+        message_content = messages[0]['content'][0]['text']
+        assert system_prompt in message_content
+        assert user_prompt in message_content
+        assert message_content.startswith(system_prompt)
+
+    @pytest.mark.asyncio
+    async def test_text_completion_streaming_chunk_count(self, text_completion_processor_streaming):
+        """Test that streaming produces expected number of chunks"""
+        # Arrange
+        system_prompt = "You are a helpful assistant."
+        user_prompt = "Test"
+
+        # Act
+        chunks = []
+        async for chunk in text_completion_processor_streaming.generate_content_stream(
+            system_prompt, user_prompt
+        ):
+            chunks.append(chunk)
+
+        # Assert
+        # Should have 15 content chunks + 1 final marker = 16 total
+        assert len(chunks) == 16
+
+        # 15 content chunks
+        content_chunks = [c for c in chunks if not c.is_final]
+        assert len(content_chunks) == 15
+
+        # 1 final marker
+        final_chunks = [c for c in chunks if c.is_final]
+        assert len(final_chunks) == 1
diff --git a/tests/test_parser_debug.py b/tests/test_parser_debug.py
new file mode 100644
index 00000000..2ad7867f
--- /dev/null
+++ b/tests/test_parser_debug.py
@@ -0,0 +1,51 @@
+"""
+Debug script to understand StreamingReActParser behavior
+"""
+import sys
+sys.path.insert(0, '/home/mark/work/trustgraph.ai/trustgraph/trustgraph-flow')
+
+from trustgraph.agent.react.streaming_parser import StreamingReActParser
+
+# Test the parser with different chunking strategies
+full_text = """Thought: I need to search for information about machine learning.
+Action: knowledge_query
+Args: {
+    "question": "What is machine learning?"
+}"""
+
+print("=" * 60)
+print("TEST 1: Send as single chunk")
+print("=" * 60)
+
+parser1 = StreamingReActParser()
+parser1.feed(full_text)
+parser1.finalize()
+result1 = parser1.get_result()
+print(f"Result: {result1}")
+print(f"Action name: '{result1.name}'")
+print(f"Arguments: {result1.arguments}")
+
+print("\n" + "=" * 60)
+print("TEST 2: Send line-by-line chunks")
+print("=" * 60)
+
+parser2 = StreamingReActParser()
+chunks = [
+    "Thought: I need to search for information about machine learning.\n",
+    "Action: knowledge_query\n",
+    "Args: {\n",
+    '    "question": "What is machine learning?"\n',
+    "}"
+]
+
+for i, chunk in enumerate(chunks):
+    print(f"\nChunk {i}: {repr(chunk)}")
+    print(f"  Before: state={parser2.state}, action_buffer='{parser2.action_buffer}'")
+    parser2.feed(chunk)
+    print(f"  After:  state={parser2.state}, action_buffer='{parser2.action_buffer}'")
+
+parser2.finalize()
+result2 = parser2.get_result()
+print(f"\nFinal Result: {result2}")
+print(f"Action name: '{result2.name}'")
+print(f"Arguments: {result2.arguments}")