`61dd56d`

fix: UI streaming improvements for multi-turn conversations

- Add ClearStream message to adapter for hiding ugly raw tool call JSON
- Add on_clear_stream handler to remove streaming widget when raw JSON detected
- Add _streamed_content flag to track if content was shown
- Finalize previous streaming widget in on_thinking_started to prevent appending
- Show fallback content in on_response_complete when nothing was streamed

Authored by

espadonne 4 months ago

SHA: 61dd56d4cbc1a2c152dac132947c5506f3183d56
Parents: 3a46b3d
Tree: ac2c4a2

2 changed files

Status	File	+	-
M	`src/loader/ui/adapter.py`	10	0
M	`src/loader/ui/app.py`	26	2

src/loader/ui/adapter.pymodified

      content: str
 +@dataclass
 +class ClearStream(Message):
 +    """Clear the current streaming content (used when raw tool calls are detected)."""
++
 +    pass
++
++
  @dataclass
  class DecompositionCreated(Message):
      """Task was decomposed into subtasks."""
                      StreamChunk(content=event.content, is_end=event.is_stream_end)
+                 )
 +            case "clear_stream":
 +                self.app.post_message(ClearStream())
++
              case "plan":
                  self.app.post_message(PlanCreated(content=event.content))

src/loader/ui/app.pymodified

  from ..agent.loop import Agent, AgentEvent
  from .adapter import (
 +    ClearStream,
      CompletionCheckPerformed,
      ConfidenceAssessed,
      CritiquePerformed,
          self.adapter = EventAdapter(self)
          self._start_time: float = 0.0
          self._current_streaming: StreamingText | None = None
 +        self._streamed_content: bool = False  # Track if any content was streamed
          self._tool_widget_queue: list[ToolCallWidget] = []  # Queue of pending tool widgets
          self._timer_handle = None
      # Message handlers from adapter
      def on_thinking_started(self, message: ThinkingStarted) -> None:
          """Handle thinking started (may be called multiple times per task)."""
 +        # Finalize any previous streaming widget to prevent appending to old content
 +        if self._current_streaming is not None:
 +            self._current_streaming.stop_streaming()
 +            self._current_streaming = None
++
 +        # Reset streamed content flag for this response iteration
 +        self._streamed_content = False
++
          # Status is already set in on_input_area_submitted, but ensure it stays on
          if not self.is_generating:
              self.is_generating = True
          self._current_streaming.append(message.content)
          msg_area.scroll_end(animate=False)
 +        # Track that we've shown actual content
 +        if message.content.strip():
 +            self._streamed_content = True
++
          if message.is_end:
              self._current_streaming.stop_streaming()
              self._current_streaming = None
 +    def on_clear_stream(self, message: ClearStream) -> None:
 +        """Clear/remove the current streaming content (used when raw tool calls are detected)."""
 +        if self._current_streaming is not None:
 +            # Remove the streaming widget entirely - it contained raw JSON tool calls
 +            self._current_streaming.remove()
 +            self._current_streaming = None
 +            self._streamed_content = False
++
      def on_tool_call_started(self, message: ToolCallStarted) -> None:
          """Handle tool call start."""
          msg_area = self.query_one("#message-area", ScrollableContainer)
      def on_response_complete(self, message: ResponseComplete) -> None:
          """Handle response completion."""
 -        # Response was already streamed, nothing extra needed
 -        pass
 +        # If no content was streamed but we have a response, display it
 +        # This handles cases like empty LLM responses with fallback messages
 +        if not self._streamed_content and message.content.strip():
 +            self._add_message(message.content)
      def on_steering_received(self, message: SteeringReceived) -> None:
          """Handle steering message being processed by agent."""