feat: include breadcrumb in embedding text

interview/rag-refactor-story.md

@@ -170,7 +170,7 @@ Known gaps:
 - Metadata taxonomy still needs cleanup, for example `database` vs `infrastructure`.
 - Abstract design questions may need query rewriting or better indexed interview/devflow documents.
 - Chunk context reconstruction is still limited when one logical section spans multiple chunks.
-- `breadcrumb` should be used more strongly in embedding text, retrieval metadata, and evidence packing.
+- `breadcrumb` now participates in embedding text, but it can still be used more strongly in context expansion, rerank, and evidence packing.
 - Rerank, RRF, BM25, and hybrid retrieval are not implemented yet.
 - Indexing writes still use SDK.
 

src/main/java/com/superbiz/agent/service/VectorIndexService.java

@@ -148,7 +148,7 @@ public class VectorIndexService {
             
             try {
                 // 生成向量
-                List<Float> vector = embeddingService.generateEmbedding(chunk.getContent());
+                List<Float> vector = embeddingService.generateEmbedding(buildEmbeddingText(chunk));
 
                 // 构建元数据（包含文件信息）
                 Map<String, Object> metadata = buildMetadata(path.toString(), chunk, chunks.size());
@@ -191,7 +191,7 @@ public class VectorIndexService {
 
             try {
                 // 生成向量
-                List<Float> vector = embeddingService.generateEmbedding(chunk.getContent());
+                List<Float> vector = embeddingService.generateEmbedding(buildEmbeddingText(chunk));
 
                 // 构建元数据（使用 docId 和 category）
                 Map<String, Object> metadata = buildDocumentMetadata(docId, chunk, chunks.size(), category);
@@ -280,6 +280,30 @@ public class VectorIndexService {
         return metadata;
     }
 
+    static String buildEmbeddingText(DocumentChunk chunk) {
+        String content = trimToEmpty(chunk.getContent());
+        String title = trimToEmpty(chunk.getTitle());
+        String breadcrumb = trimToEmpty(chunk.getBreadcrumb());
+
+        if (title.isEmpty() && breadcrumb.isEmpty()) {
+            return content;
+        }
+
+        StringBuilder text = new StringBuilder();
+        if (!title.isEmpty()) {
+            text.append("Title: ").append(title).append("\n");
+        }
+        if (!breadcrumb.isEmpty()) {
+            text.append("Path: ").append(breadcrumb).append("\n");
+        }
+        text.append("Content:\n").append(content);
+        return text.toString();
+    }
+
+    private static String trimToEmpty(String value) {
+        return value == null ? "" : value.trim();
+    }
+
     /**
      * 删除文件的旧数据（根据 metadata._source）
      */

src/test/java/com/superbiz/agent/service/VectorIndexServiceTest.java

@@ -0,0 +1,37 @@
+package com.superbiz.agent.service;
+
+import com.superbiz.agent.dto.DocumentChunk;
+import org.junit.jupiter.api.Test;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+class VectorIndexServiceTest {
+
+    @Test
+    void buildEmbeddingTextIncludesTitleAndBreadcrumb() {
+        DocumentChunk chunk = DocumentChunk.builder()
+                .title("Connection Pool")
+                .breadcrumb("Database > MySQL > Connection Pool")
+                .content("Check active connections and leak detection.")
+                .build();
+
+        String embeddingText = VectorIndexService.buildEmbeddingText(chunk);
+
+        assertEquals("""
+                Title: Connection Pool
+                Path: Database > MySQL > Connection Pool
+                Content:
+                Check active connections and leak detection.""", embeddingText);
+    }
+
+    @Test
+    void buildEmbeddingTextKeepsPlainContentWhenNoStructureExists() {
+        DocumentChunk chunk = DocumentChunk.builder()
+                .title(" ")
+                .breadcrumb(null)
+                .content("Plain chunk content.")
+                .build();
+
+        assertEquals("Plain chunk content.", VectorIndexService.buildEmbeddingText(chunk));
+    }
+}