feat: implement error handling for empty upstream responses in chat streams and update UI to display stream-level errors

internal/adapter/openai/chat_stream_runtime.go

@@ -98,6 +98,19 @@ func (s *chatStreamRuntime) sendDone() {
 	}
 }
 
+func (s *chatStreamRuntime) sendFailedChunk(status int, message, code string) {
+	s.sendChunk(map[string]any{
+		"status_code": status,
+		"error": map[string]any{
+			"message": message,
+			"type":    openAIErrorType(status),
+			"code":    code,
+			"param":   nil,
+		},
+	})
+	s.sendDone()
+}
+
 func (s *chatStreamRuntime) finalize(finishReason string) {
 	finalThinking := s.thinking.String()
 	finalText := cleanVisibleOutput(s.text.String(), s.stripReferenceMarkers)
@@ -168,6 +181,21 @@ func (s *chatStreamRuntime) finalize(finishReason string) {
 	if len(detected.Calls) > 0 || s.toolCallsEmitted {
 		finishReason = "tool_calls"
 	}
+	if len(detected.Calls) == 0 && !s.toolCallsEmitted && strings.TrimSpace(finalText) == "" {
+		status := http.StatusTooManyRequests
+		message := "Upstream model returned empty output."
+		code := "upstream_empty_output"
+		if strings.TrimSpace(finalThinking) != "" {
+			message = "Upstream model returned reasoning without visible output."
+		}
+		if finishReason == "content_filter" {
+			status = http.StatusBadRequest
+			message = "Upstream content filtered the response and returned no output."
+			code = "content_filter"
+		}
+		s.sendFailedChunk(status, message, code)
+		return
+	}
 	usage := openaifmt.BuildChatUsage(s.finalPrompt, finalThinking, finalText)
 	s.sendChunk(openaifmt.BuildChatStreamChunk(
 		s.completionID,
@@ -184,6 +212,9 @@ func (s *chatStreamRuntime) onParsed(parsed sse.LineResult) streamengine.ParsedD
 		return streamengine.ParsedDecision{}
 	}
 	if parsed.ContentFilter {
+		if strings.TrimSpace(s.text.String()) == "" {
+			return streamengine.ParsedDecision{Stop: true, StopReason: streamengine.StopReason("content_filter")}
+		}
 		return streamengine.ParsedDecision{Stop: true, StopReason: streamengine.StopReasonHandlerRequested}
 	}
 	if parsed.ErrorMessage != "" {

internal/adapter/openai/stream_status_test.go

@@ -243,6 +243,49 @@ func TestChatCompletionsStreamContentFilterStopsNormallyWithoutLeak(t *testing.T
 	}
 }
 
+func TestChatCompletionsStreamEmitsFailureFrameWhenUpstreamOutputEmpty(t *testing.T) {
+	statuses := make([]int, 0, 1)
+	h := &Handler{
+		Store: mockOpenAIConfig{wideInput: true},
+		Auth:  streamStatusAuthStub{},
+		DS:    streamStatusDSStub{resp: makeOpenAISSEHTTPResponse("data: [DONE]")},
+	}
+	r := chi.NewRouter()
+	r.Use(captureStatusMiddleware(&statuses))
+	RegisterRoutes(r, h)
+
+	reqBody := `{"model":"deepseek-chat","messages":[{"role":"user","content":"hi"}],"stream":true}`
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(reqBody))
+	req.Header.Set("Authorization", "Bearer direct-token")
+	req.Header.Set("Content-Type", "application/json")
+	rec := httptest.NewRecorder()
+	r.ServeHTTP(rec, req)
+
+	if rec.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d body=%s", rec.Code, rec.Body.String())
+	}
+	if len(statuses) != 1 || statuses[0] != http.StatusOK {
+		t.Fatalf("expected captured status 200, got %#v", statuses)
+	}
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	if len(frames) != 1 {
+		t.Fatalf("expected one failure frame, got %#v body=%s", frames, rec.Body.String())
+	}
+	last := frames[0]
+	statusCode, ok := last["status_code"].(float64)
+	if !ok || int(statusCode) != http.StatusTooManyRequests {
+		t.Fatalf("expected status_code=429, got %#v body=%s", last["status_code"], rec.Body.String())
+	}
+	errObj, _ := last["error"].(map[string]any)
+	if asString(errObj["code"]) != "upstream_empty_output" {
+		t.Fatalf("expected code=upstream_empty_output, got %#v", last)
+	}
+}
+
 func TestResponsesStreamUsageIgnoresBatchAccumulatedTokenUsage(t *testing.T) {
 	statuses := make([]int, 0, 1)
 	h := &Handler{