修复搜索场景 citation 偶发未替换

internal/sse/consumer.go

@@ -29,6 +29,7 @@ func CollectStream(resp *http.Response, thinkingEnabled bool, closeBody bool) Co
 	text := strings.Builder{}
 	thinking := strings.Builder{}
 	contentFilter := false
+	stopped := false
 	collector := newCitationLinkCollector()
 	currentType := "text"
 	if thinkingEnabled {
@@ -38,6 +39,9 @@ func CollectStream(resp *http.Response, thinkingEnabled bool, closeBody bool) Co
 		if chunk, done, parsed := ParseDeepSeekSSELine(line); parsed && !done {
 			collector.ingestChunk(chunk)
 		}
+		if stopped {
+			return true
+		}
 		result := ParseDeepSeekContentLine(line, thinkingEnabled, currentType)
 		currentType = result.NextType
 		if !result.Parsed {
@@ -47,7 +51,10 @@ func CollectStream(resp *http.Response, thinkingEnabled bool, closeBody bool) Co
 			if result.ContentFilter {
 				contentFilter = true
 			}
-			return false
+			// Keep scanning to collect late-arriving citation metadata lines
+			// that can appear after response/status=FINISHED.
+			stopped = true
+			return true
 		}
 		for _, p := range result.Parts {
 			if p.Type == "thinking" {

internal/sse/consumer_edge_test.go

@@ -185,6 +185,24 @@ func TestCollectStreamExtractsCitationLinksWithRepeatedURLsAndNilIndices(t *test
 	}
 }
 
+func TestCollectStreamCollectsCitationLinksAfterFinished(t *testing.T) {
+	resp := makeHTTPResponse(
+		"data: {\"p\":\"response/content\",\"v\":\"结论[citation:1]\"}\n" +
+			"data: {\"p\":\"response/status\",\"v\":\"FINISHED\"}\n" +
+			"data: {\"p\":\"response/fragments/-1/results\",\"v\":[{\"url\":\"https://example.com/a\",\"cite_index\":1}]}\n" +
+			"data: {\"p\":\"response/content\",\"v\":\"should-not-append\"}\n" +
+			"data: [DONE]\n",
+	)
+
+	result := CollectStream(resp, false, false)
+	if result.Text != "结论[citation:1]" {
+		t.Fatalf("expected text to freeze after finished, got %q", result.Text)
+	}
+	if got := result.CitationLinks[1]; got != "https://example.com/a" {
+		t.Fatalf("expected citation 1 link, got %q", got)
+	}
+}
+
 func TestCollectStreamMultipleThinkingChunks(t *testing.T) {
 	resp := makeHTTPResponse(
 		"data: {\"p\":\"response/thinking_content\",\"v\":\"part1\"}\n" +