refactor: replace bufio.Scanner with bufio.Reader for SSE stream parsing and track emitted text to prevent redundant output blocks

internal/httpapi/claude/handler_stream_test.go

@@ -28,6 +28,18 @@ func makeClaudeSSEHTTPResponse(lines ...string) *http.Response {
 	}
 }
 
+func makeClaudeContentLine(t *testing.T, text string) string {
+	t.Helper()
+	line, err := json.Marshal(map[string]any{
+		"p": "response/content",
+		"v": text,
+	})
+	if err != nil {
+		t.Fatalf("marshal content line failed: %v", err)
+	}
+	return "data: " + string(line)
+}
+
 func parseClaudeFrames(t *testing.T, body string) []claudeFrame {
 	t.Helper()
 	chunks := strings.Split(body, "\n\n")
@@ -71,6 +83,17 @@ func findClaudeFrames(frames []claudeFrame, event string) []claudeFrame {
 	return out
 }
 
+func collectClaudeTextDeltas(frames []claudeFrame) string {
+	var combined strings.Builder
+	for _, f := range findClaudeFrames(frames, "content_block_delta") {
+		delta, _ := f.Payload["delta"].(map[string]any)
+		if delta["type"] == "text_delta" {
+			combined.WriteString(asString(delta["text"]))
+		}
+	}
+	return combined.String()
+}
+
 func TestHandleClaudeStreamRealtimeTextIncrementsWithEventHeaders(t *testing.T) {
 	h := &Handler{}
 	resp := makeClaudeSSEHTTPResponse(
@@ -111,6 +134,26 @@ func TestHandleClaudeStreamRealtimeTextIncrementsWithEventHeaders(t *testing.T)
 	}
 }
 
+func TestHandleClaudeStreamRealtimeToolBufferedPlainTextDoesNotRepeatFinalText(t *testing.T) {
+	h := &Handler{}
+	want := "明白\n\nBash\nIN\npwd\nOUT\nok"
+	resp := makeClaudeSSEHTTPResponse(
+		makeClaudeContentLine(t, "明"),
+		makeClaudeContentLine(t, "白\n\nBash\nIN\npwd\n"),
+		makeClaudeContentLine(t, "OUT\nok"),
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/anthropic/v1/messages", nil)
+
+	h.handleClaudeStreamRealtime(rec, req, resp, "claude-sonnet-4-5", []any{map[string]any{"role": "user", "content": "use tool"}}, false, false, []string{"Bash"}, nil)
+
+	frames := parseClaudeFrames(t, rec.Body.String())
+	if got := collectClaudeTextDeltas(frames); got != want {
+		t.Fatalf("unexpected combined text: got %q want %q body=%s", got, want, rec.Body.String())
+	}
+}
+
 func TestHandleClaudeStreamRealtimeTrimsContinuationReplay(t *testing.T) {
 	h := &Handler{}
 	prefix := strings.Repeat("A", 40)

internal/httpapi/claude/stream_runtime_core.go

@@ -43,6 +43,7 @@ type claudeStreamRuntime struct {
 	thinkingBlockIndex int
 	textBlockOpen      bool
 	textBlockIndex     int
+	textEmitted        bool
 	ended              bool
 	upstreamErr        string
 }
@@ -181,6 +182,7 @@ func (s *claudeStreamRuntime) onParsed(parsed sse.LineResult) streamengine.Parse
 					"text": cleanedText,
 				},
 			})
+			s.textEmitted = true
 			continue
 		}
 
@@ -226,6 +228,7 @@ func (s *claudeStreamRuntime) onParsed(parsed sse.LineResult) streamengine.Parse
 					"text": cleaned,
 				},
 			})
+			s.textEmitted = true
 		}
 	}
 

internal/httpapi/claude/stream_runtime_finalize.go

@@ -109,6 +109,7 @@ func (s *claudeStreamRuntime) finalize(stopReason string) {
 						"text": cleaned,
 					},
 				})
+				s.textEmitted = true
 			}
 		}
 	}
@@ -141,7 +142,7 @@ func (s *claudeStreamRuntime) finalize(stopReason string) {
 				s.nextBlockIndex++
 				s.sendToolUseBlock(idx, tc)
 			}
-		} else if finalText != "" && !s.textBlockOpen {
+		} else if finalText != "" && !s.textEmitted {
 			idx := s.nextBlockIndex
 			s.nextBlockIndex++
 			s.send("content_block_start", map[string]any{
@@ -160,6 +161,7 @@ func (s *claudeStreamRuntime) finalize(stopReason string) {
 					"text": finalText,
 				},
 			})
+			s.textEmitted = true
 			s.send("content_block_stop", map[string]any{
 				"type":  "content_block_stop",
 				"index": idx,