In the streaming path, sentry_sdk.traces.continue_trace(headers) is never called before start_span, so incoming sentry-trace/traceparent/baggage headers from HTTP requests are never propagated — every Chalice invocation starts a fresh, unconnected trace instead of joining the upstream trace. Add sentry_sdk.traces.continue_trace(headers) before the start_span call (mirroring the GCP integration at gcp.py:133).

In sentry_sdk/integrations/chalice.py, _get_lambda_span_attributes calls invoked_arn.split(':') without a null guard. In real AWS Lambda the runtime always populates invoked_function_arn, but chalice.app.LambdaContext (used by chalice local) leaves it as None by default. When span streaming is enabled (trace_lifecycle=stream) and running under chalice local, both streaming paths (EventSourceHandler.__call__ and _get_view_function_response.wrapped_view_function) call _get_lambda_span_attributes outside capture_internal_exceptions, so the resulting AttributeError: 'NoneType' object has no attribute 'split' propagates and crashes the handler. Production AWS Lambda is unaffected. Fix by defaulting to an empty string: invoked_arn = aws_context.invoked_function_arn or ''.