stream responses and all tests passing

agent/loop.py

@@ -10,7 +10,21 @@ from agent.tools import TOOL_SCHEMAS, dispatch_tool
 client = AsyncAnthropic(api_key=settings.anthropic_api_key)
 
 
-async def run_turn(user_message: str, history: list[dict] = None, sandbox=None) -> str:
+async def run_turn(
+    user_message: str, history: list[dict] = None, sandbox=None, stream_callback=None
+) -> str:
+    """
+    Run one turn of agent loop with streaming support.
+
+    Args:
+        user_message: User's input
+        history: Conversation history
+        sandbox: Sandbox session
+        stream_callback: Optional callback(text) for streaming output
+
+    Returns:
+        Final text response
+    """
 
     if history is None:
         history = []
@@ -18,44 +32,76 @@ async def run_turn(user_message: str, history: list[dict] = None, sandbox=None)
     # add the new user message to history
     messages = history + [{"role": "user", "content": user_message}]
 
-    response = await client.messages.create(
+    full_response = ""  # accumulate full response from stream
+
+    async with client.messages.stream(
         model=settings.model,
         max_tokens=settings.max_tokens,
         tools=TOOL_SCHEMAS,
         messages=messages,
-    )
+    ) as stream:
+        # stream text as it arrives
+        async for text in stream.text_stream:
+            full_response += text
+            if stream_callback:
+                await stream_callback(text)
+
+        # get the final message (includes tool calls)
+        response = await stream.get_final_message()
 
     while response.stop_reason == "tool_use":
         tool_results = []
         for block in response.content:
             if block.type == "tool_use":
+                # show what tool is being called
+                if stream_callback:
+                    await stream_callback(
+                        f"\n🔧 Running: {block.name}({block.input})\n"
+                    )
+
+                # Execute tool
                 result = await dispatch_tool(
                     tool_name=block.name, tool_input=block.input, sandbox=sandbox
                 )
+
+                # show result preview
+                if stream_callback:
+                    preview = result[:100] + "..." if len(result) > 100 else result
+                    await stream_callback(f"✓ Result: {preview}\n\n")
+
                 tool_results.append(
                     {"type": "tool_result", "tool_use_id": block.id, "content": result}
                 )
 
+        # update messages with tool results
         messages = messages + [
             {"role": "assistant", "content": response.content},
             {"role": "user", "content": tool_results},
         ]
 
-        response = await client.messages.create(
+        # get next response
+        async with client.messages.stream(
             model=settings.model,
             max_tokens=settings.max_tokens,
             tools=TOOL_SCHEMAS,
             messages=messages,
-        )
+        ) as stream:
+            async for text in stream.text_stream:
+                full_response += text
+                if stream_callback:
+                    await stream_callback(text)
 
-    return next(block.text for block in response.content if hasattr(block, "text"))
+            response = await stream.get_final_message()
+
+    return full_response
 
 
 async def run_session(sandbox=None):
     """simple CLI session - temporary until TUI is built"""
     history = ConversationHistory()
 
-    print("Codeing agent ready. Type /quit to quit.")
+    print(f"Codeing agent ready. Session: {history.session_id}")
+    print("Type /quit to quit.\n")
 
     while True:
         user_input = input("You: ").strip()
@@ -71,4 +117,19 @@ async def run_session(sandbox=None):
         response = await run_turn(user_input, history.get_all(), sandbox)
         history.add_message("assistant", response)
 
-        print(f"\nAssistant: {response}")
+        # Print "Agent: " then stream response
+        print(f"\nAgent: ", end="", flush=True)
+
+        # callback that prints text as it arrives
+        async def print_stream(text: str):
+            print(text, end="", flush=True)
+
+        # run turn with streaming
+        response = await run_turn(
+            user_input, history.get_all(), sandbox, stream_callback=print_stream
+        )
+
+        print()  # new line after response
+
+        # add to history
+        history.add_message("assistant", response)

sandbox/session.py

@@ -30,12 +30,12 @@ class PodmanSandbox:
         )
         return self
 
-    async def run(self, command: str) -> str:
+    def run(self, command: str) -> str:
         """Execute command in microVM/"""
         exit_code, output = self.container.exec_run(
-            ["/bin/sh", "-c", command], workdir="/workspace"
+            ["/bin/sh", "-c", command], workdir="/workspace", demux=False
         )
-        return output.decode()
+        return output.decode("utf-8", errors="replace")
 
     async def __aexit__(self, *args):
         if self.container:

tests/conftest.py

@@ -24,24 +24,42 @@ def settings():
     )
 
 
+# tests/conftest.py - replace mock_anthropic_client fixture
+
+
 @pytest.fixture
 def mock_anthropic_client():
-    """Mock anthropic client that returns a fake response."""
+    """Mock Anthropic client with streaming support."""
     mock_client = AsyncMock()
 
-    # create a realistic fake response
+    # Create mock stream
-    fake_message = Message(
+    mock_stream = AsyncMock()
-        id="msg_test123",
+
-        type="message",
+    # Mock text_stream - simple async generator
-        role="assistant",
+    async def fake_text():
-        content=[TextBlock(type="text", text="42")],
+        yield "42"
-        model="claude-test-model",
+
-        stop_reason="end_turn",
+    mock_stream.text_stream = fake_text()
-        usage=Usage(input_tokens=10, output_tokens=5),
+
+    # Mock get_final_message
+    mock_stream.get_final_message = AsyncMock(
+        return_value=Message(
+            id="msg_test123",
+            type="message",
+            role="assistant",
+            content=[TextBlock(type="text", text="42")],
+            model="claude-test-model",
+            stop_reason="end_turn",
+            usage=Usage(input_tokens=10, output_tokens=1),
+        )
     )
 
-    # make messages.create() return this fake message
+    # Mock context manager
-    mock_client.messages.create = AsyncMock(return_value=fake_message)
+    mock_stream.__aenter__ = AsyncMock(return_value=mock_stream)
+    mock_stream.__aexit__ = AsyncMock()
+
+    # Wire up
+    mock_client.messages.stream = MagicMock(return_value=mock_stream)
 
     return mock_client
 

tests/test_loop.py

@@ -1,4 +1,4 @@
-from unittest.mock import AsyncMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
@@ -16,13 +16,13 @@ async def test_run_turn_basic(mock_anthropic_client):
         result = await run_turn("What is 2+2?")
 
     # verify client was called
-    mock_anthropic_client.messages.create.assert_called_once()
+    mock_anthropic_client.messages.stream.assert_called_once()
 
     # verify message returned
     assert result == "42"
 
     # verify call has correct parameters
-    call_args = mock_anthropic_client.messages.create.call_args
+    call_args = mock_anthropic_client.messages.stream.call_args
 
 
 @pytest.mark.unit
@@ -34,7 +34,7 @@ async def test_run_turn_with_history(mock_anthropic_client, sample_history):
     with patch("agent.loop.client", mock_anthropic_client):
         result = await run_turn("What is 2+2?", history=sample_history)
 
-    call_args = mock_anthropic_client.messages.create.call_args
+    call_args = mock_anthropic_client.messages.stream.call_args
     messages = call_args.kwargs["messages"]
 
     # verify all history was included plus new message
@@ -53,7 +53,7 @@ async def test_run_turn_uses_settings(mock_anthropic_client, settings):
             await run_turn("test message")
 
     # Verify settings were used
-    call_args = mock_anthropic_client.messages.create.call_args
+    call_args = mock_anthropic_client.messages.stream.call_args
     assert call_args.kwargs["model"] == settings.model
     assert call_args.kwargs["max_tokens"] == settings.max_tokens
 
@@ -63,16 +63,18 @@ async def test_run_turn_uses_settings(mock_anthropic_client, settings):
 async def test_run_session_calls_run_turn_with_user_input():
     """Test that user input is passed to run_turn"""
 
-    # Mock the input()
+    mock_history = MagicMock()
-    with patch("builtins.input", side_effect=["hello", KeyboardInterrupt]):
+    mock_history.get_all.return_value = []
-        # Mock run_turn to avoid actually calling the API
+    mock_history.session_id = "test-session"
-        with patch("agent.loop.run_turn", new_callable=AsyncMock) as mock_run_turn:
-            mock_run_turn.return_value = AsyncMock(content=[AsyncMock(text="response")])
 
-            try:
+    with patch("agent.loop.ConversationHistory", return_value=mock_history):
+        with patch("agent.loop.run_turn", new_callable=AsyncMock) as mock_run_turn:
+            mock_run_turn.return_value = "response text"
+
+            # Provide inputs: command, then quit
+            inputs = iter(["hello", "/quit"])
+            with patch("builtins.input", side_effect=lambda _: next(inputs)):
                 await run_session()
-            except KeyboardInterrupt:
-                pass
 
     mock_run_turn.assert_called()
     assert mock_run_turn.call_args.args[0] == "hello"

tests/test_sandbox.py

@@ -67,11 +67,13 @@ async def test_sandbox_run_executes_command():
         sb = PodmanSandbox()
         await sb.__aenter__()
 
-        result = await sb.run("echo 'hello from sandbox'")
+        result = sb.run("echo 'hello from sandbox'")
 
         # Verify exec_run was called with shell wrapper
         mock_container.exec_run.assert_called_once_with(
-            ["/bin/sh", "-c", "echo 'hello from sandbox'"], workdir="/workspace"
+            ["/bin/sh", "-c", "echo 'hello from sandbox'"],
+            workdir="/workspace",
+            demux=False,
         )
         assert result == "hello from sandbox\n"
 
@@ -83,7 +85,7 @@ async def test_tool_call_fails_if_sandbox_crashes():
 
     # Simulate crashed sandbox (container is None)
     mock_sandbox = MagicMock()
-    mock_sandbox.run = AsyncMock(side_effect=RuntimeError("Container crashed"))
+    mock_sandbox.run = MagicMock(side_effect=RuntimeError("Container crashed"))
 
     result = await bash("ls -la", mock_sandbox)
 

tools/bash.py

@@ -1,3 +1,6 @@
+import asyncio
+
+
 async def bash(command: str, sandbox=None) -> str:
     """
     Execute a bash command in the sandbox.
@@ -14,7 +17,7 @@ async def bash(command: str, sandbox=None) -> str:
         return "Error: Sandbox not available"
 
     try:
-        result = await sandbox.run(command)
+        result = await asyncio.to_thread(sandbox.run, command)
         return result
 
     except RuntimeError as e: