fix(streaming): normalize finish_reason in get_result() per streaming…

amplifier_module_provider_github_copilot/streaming.py

@@ -251,14 +251,27 @@ def add(self, event: DomainEvent) -> None:
             self.is_complete = True
 
     def get_result(self) -> AccumulatedResponse:
-        """Get accumulated response."""
+        """Get accumulated response.
+
+        Applies the same finish_reason normalization as to_chat_response() so that
+        callers of either method see canonical values per streaming-contract:FinishReason:MUST:5.
+        """
+        # streaming-contract:FinishReason:MUST:5 — normalize to lowercase canonical values
+        if self.tool_calls:
+            # Tool calls always override SDK finish_reason (deny/capture flow)
+            normalized_finish_reason: str | None = "tool_calls"
+        elif self.finish_reason:
+            normalized_finish_reason = self.finish_reason.lower()
+        else:
+            normalized_finish_reason = self.finish_reason
+
         return AccumulatedResponse(
             text_content=self.text_content,
             thinking_content=self.thinking_content,
             reasoning_opaque=self.reasoning_opaque,
             tool_calls=self.tool_calls,
             usage=self.usage,
-            finish_reason=self.finish_reason,
+            finish_reason=normalized_finish_reason,
             error=self.error,
             is_complete=self.is_complete,
         )
@@ -381,8 +394,11 @@ def to_chat_response(self) -> "StreamingChatResponse":
             # Use "stop" per amplifier-core proto (not "end_turn" which is an SDK input key)
             normalized_finish_reason = "stop"
         else:
-            # No tool calls but SDK provided finish_reason: preserve it
-            normalized_finish_reason = self.finish_reason
+            # No tool calls but SDK provided finish_reason: normalize to lowercase
+            # streaming-contract:FinishReason:MUST:5 — finish_reason MUST be lowercase
+            # Defense-in-depth: translate_event() applies finish_reason_map first, but
+            # the accumulator normalizes here in case raw SDK values pass through directly.
+            normalized_finish_reason = self.finish_reason.lower()
 
         # Contract: content_blocks is None when empty (not empty list)
         # streaming-contract:StreamingResponse:MUST:4

tests/conftest.py

@@ -245,22 +245,6 @@ def sdk_module() -> Any:
     return pytest.importorskip("copilot", reason="github-copilot-sdk not installed")
 
 
-@pytest.fixture
-def restore_github_tokens() -> Generator[None, None, None]:
-    """Restore saved GitHub tokens for live tests.
-
-    Use this fixture in @pytest.mark.live tests that need real tokens.
-    Tokens were saved and cleared at module load for security (P1-6).
-    """
-    for var, val in _saved_tokens.items():
-        os.environ[var] = val
-    yield
-    # Restore cleared state after test
-    for var in _saved_tokens:
-        if var in os.environ:
-            del os.environ[var]
-
-
 @pytest.fixture(autouse=True)
 def _auto_restore_tokens_for_live_tests(  # pyright: ignore[reportUnusedFunction]
     request: pytest.FixtureRequest,
@@ -284,44 +268,3 @@ def _auto_restore_tokens_for_live_tests(  # pyright: ignore[reportUnusedFunction
         yield
 
 
-@pytest.fixture
-def mock_sdk_event_dict() -> dict[str, Any]:
-    """Sample SDK event as dict for testing helpers.
-
-    Contract: sdk-boundary:EventShape:MUST:1
-    Reference: SDK SessionEvent structure from github-copilot-sdk
-
-    SDK v0.1.33+ uses nested data structure:
-    - event.data.delta_content for streaming deltas
-    - event.data.content for complete messages
-    """
-    return {
-        "type": "assistant.message_delta",
-        "data": {
-            "delta_content": "hello",
-            "message_id": "msg_001",
-        },
-    }
-
-
-@pytest.fixture
-def mock_sdk_event_object() -> Any:
-    """Sample SDK event as object for testing helpers.
-
-    Contract: sdk-boundary:EventShape:MUST:2, sdk-boundary:EventShape:MUST:3
-    Reference: SDK SessionEvent structure from github-copilot-sdk
-
-    Matches real SDK SessionEvent structure from generated SessionEvents.
-    """
-
-    class MockData:
-        delta_content = "hello"
-        content = None
-        message_id = "msg_001"
-        reasoning_id = None
-
-    class MockEvent:
-        type = "assistant.message_delta"
-        data = MockData()
-
-    return MockEvent()

tests/fixtures/__init__.py

@@ -12,7 +12,6 @@
     SessionEventType,
     error_event,
     idle_event,
-    message_complete_event,
     text_delta_event,
     usage_event,
 )
@@ -33,7 +32,6 @@
     "SessionEventType",
     "idle_event",
     "text_delta_event",
-    "message_complete_event",
     "error_event",
     "usage_event",
     # SDK response fixtures

tests/fixtures/sdk_mocks.py

@@ -289,14 +289,6 @@ def text_delta_event(text: str) -> SessionEvent:
     )
 
 
-def message_complete_event(finish_reason: str = "stop") -> SessionEvent:
-    """Create message complete event."""
-    return SessionEvent(
-        type=SessionEventType.ASSISTANT_MESSAGE,
-        data=SessionEventData(finish_reason=finish_reason),
-    )
-
-
 def error_event(message: str) -> SessionEvent:
     """Create error event."""
     return SessionEvent(

tests/fixtures/sdk_responses.py

@@ -61,35 +61,5 @@ class MockUsage:
     total_tokens: int = 0
 
 
-# Pre-built fixtures for common test scenarios
-
-
-def create_simple_response(content: str) -> MockSDKResponse:
-    """Create a simple SDK response with text content."""
-    return MockSDKResponse(data=MockData(content=content))
-
-
-def create_empty_response() -> MockSDKResponse:
-    """Create an SDK response with empty content."""
-    return MockSDKResponse(data=MockData(content=""))
-
-
-def create_none_response() -> MockSDKResponse:
-    """Create an SDK response with None data."""
-    return MockSDKResponse(data=None)
-
-
-def create_dict_response(content: str) -> MockSDKResponse:
-    """Create an SDK response with dict data (backward compat)."""
-    return MockSDKResponse(data={"content": content})
-
-
-# Fixture registry for parametrized tests
-SDK_RESPONSE_FIXTURES = {
-    "simple": create_simple_response("Hello, world!"),
-    "empty": create_empty_response(),
-    "none": create_none_response(),
-    "dict": create_dict_response("Dict content"),
-    "multiline": create_simple_response("Line 1\nLine 2\nLine 3"),
-    "unicode": create_simple_response("Unicode: 你好世界 🌍"),
-}
+# Pre-built fixtures for common test scenarios and SDK_RESPONSE_FIXTURES removed —
+# zero consumers found (orphaned from deleted test files).