feat(rag): close eval pipeline with live snapshots

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

@@ -126,11 +126,13 @@ public class DocumentManagementService {
             // 5. 解析 frontmatter
             long frontmatterStart = System.currentTimeMillis();
             Frontmatter frontmatter = null;
+            String bodyText = text;
             if (frontmatterParser.hasFrontmatter(text)) {
                 frontmatter = frontmatterParser.parse(text);
                 if (frontmatter != null) {
                     // LLM 补全 covers / whenToRetrieve（已有值则跳过）
-                    documentFieldEnricher.enrich(frontmatter, text, category);
+                    bodyText = frontmatterParser.stripFrontmatter(text);
+                    documentFieldEnricher.enrich(frontmatter, bodyText, category);
                     log.info("解析到frontmatter: title={}, keywords={}, time={}ms",
                         frontmatter.getTitle(), frontmatter.getKeywords(), System.currentTimeMillis() - frontmatterStart);
                 } else {
@@ -142,7 +144,7 @@ public class DocumentManagementService {
 
             // 6. 分块
             long chunkStart = System.currentTimeMillis();
-            List<DocumentChunk> chunks = documentChunkService.chunkDocument(text, fileName);
+            List<DocumentChunk> chunks = documentChunkService.chunkDocument(bodyText, fileName);
             if (chunks.isEmpty()) {
                 throw new DocumentProcessException(fileName, "upload", "文档分块失败");
             }
@@ -150,7 +152,7 @@ public class DocumentManagementService {
                 fileName, chunks.size(), System.currentTimeMillis() - chunkStart);
 
             // 7. 创建文档元数据
-            String docId = UUID.randomUUID().toString();
+            String docId = resolveDocumentId(frontmatter);
             String metadataJson = null;
             if (frontmatter != null) {
                 try {
@@ -181,7 +183,7 @@ public class DocumentManagementService {
             // 8. 向量化并索引
             try {
                 long vectorStart = System.currentTimeMillis();
-                vectorIndexService.indexDocumentChunks(docId, chunks, category);
+                vectorIndexService.indexDocumentChunks(docId, chunks, category, frontmatter);
                 document.setStatus("INDEXED");
                 document.setIndexedAt(LocalDateTime.now());
                 apiDocumentRepository.save(document);
@@ -203,6 +205,7 @@ public class DocumentManagementService {
                     .keywords(frontmatter.getKeywords())
                     .summary(frontmatter.getSummary())
                     .category(category)
+                    .kbScope(frontmatter.getKbScope())
                     .sections(frontmatter.getSections())
                     .covers(frontmatter.getCovers())
                     .whenToRetrieve(frontmatter.getWhenToRetrieve())
@@ -311,6 +314,16 @@ public class DocumentManagementService {
         }
     }
 
+    private String resolveDocumentId(Frontmatter frontmatter) {
+        if (frontmatter != null && frontmatter.getSource() != null) {
+            String source = frontmatter.getSource().trim();
+            if (!source.isEmpty() && source.length() <= 64) {
+                return source;
+            }
+        }
+        return UUID.randomUUID().toString();
+    }
+
     /**
      * 根据 docId 查询文档
      */