feat: integrate RAG rag_jrxml submodule and fix Anthropic API key

Add rag submodule for semantic JRXML chunk retrieval, refactor
retrieve node to use RAGSearcher, and fix missing api_key in
Anthropic SDK client initialization.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

scripts/init_kb.py

@@ -1,120 +1,55 @@
-"""初始化 Chroma 知识库，加载示例 JRXML 模板和错误修正案例。
+"""初始化 JRXML 向量知识库。
 
-用法: python scripts/init_kb.py
+rag_jrxml 子项目独立运行管线（分块→向量化→导入），本脚本仅用于预下载嵌入模型。
+
+用法:
+    python scripts/init_kb.py --download-model   # 预下载嵌入模型
 """
 
 import os
 import sys
+import argparse
 from pathlib import Path
 
 from dotenv import load_dotenv
 
 sys.path.insert(0, str(Path(__file__).parent.parent))
-
 load_dotenv()
 
 
-def download_embeddings_model():
-    """预下载 Qwen3-Embedding 模型（从 HuggingFace）。
-
-    用法: python scripts/init_kb.py --download-model
-    """
-    model_name = os.getenv("LOCAL_EMBED_MODEL", "Qwen/Qwen3-Embedding-0.6B")
+def download_model():
+    """预下载嵌入模型到本地。"""
+    model_name = os.getenv("RAG_EMBED_MODEL", "Qwen/Qwen3-Embedding-0.6B")
     print(f"正在下载嵌入模型: {model_name}")
-    print("如遇网络超时，可手动执行以下命令后重试:")
-    print(f"  huggingface-cli download {model_name} --local-dir ./models/{model_name.replace('/', '_')}")
+    print("如遇网络超时，可设置环境变量 HF_ENDPOINT=https://hf-mirror.com 使用镜像")
     print()
 
-    try:
-        from langchain_huggingface import HuggingFaceEmbeddings
-    except ImportError:
-        print("错误: 请先安装 huggingface 依赖")
-        print("  pip install langchain-huggingface sentence-transformers")
-        return
+    from sentence_transformers import SentenceTransformer
 
-    # HuggingFaceEmbeddings 会在首次调用时自动下载模型
-    embeddings = HuggingFaceEmbeddings(model_name=model_name)
-    # 调用一次以确保完全下载
-    embeddings.embed_query("测试")
+    model = SentenceTransformer(model_name)
+    model.encode("测试下载")
     print(f"嵌入模型下载完成: {model_name}")
 
-from backend.embeddings import get_embeddings
-
-
-def load_templates(template_dir: Path) -> list[dict]:
-    docs = []
-    for fpath in template_dir.glob('*.jrxml'):
-        content = fpath.read_text(encoding='utf-8')
-        name = fpath.stem
-        docs.append({
-            'content': content,
-            'metadata': {
-                'source': str(fpath),
-                'type': 'full_report',
-                'name': name,
-            },
-        })
-    return docs
-
-
-def load_corrections(corrections_dir: Path) -> list[dict]:
-    docs = []
-    for fpath in corrections_dir.glob('*.jrxml'):
-        content = fpath.read_text(encoding='utf-8')
-        docs.append({
-            'content': content,
-            'metadata': {
-                'source': str(fpath),
-                'type': 'correction_case',
-                'name': fpath.stem,
-            },
-        })
-    return docs
-
 
 def main():
-    persist_dir = os.getenv('CHROMA_PERSIST_DIR', './db/chroma')
-    data_dir = Path(__file__).parent.parent / 'data'
-
-    template_dir = data_dir / 'sample_templates'
-    corrections_dir = data_dir / 'corrections'
-
-    docs = []
-    if template_dir.exists():
-        docs.extend(load_templates(template_dir))
-        print(f'从 {template_dir} 加载了 {len(docs)} 个模板')
-
-    if corrections_dir.exists():
-        corr = load_corrections(corrections_dir)
-        docs.extend(corr)
-        print(f'从 {corrections_dir} 加载了 {len(corr)} 个修正案例')
-
-    if not docs:
-        print('未找到文档，无需索引。')
-        return
-
-    embeddings = get_embeddings()
-    from langchain_chroma import Chroma
-
-    texts = [d['content'] for d in docs]
-    metadatas = [d['metadata'] for d in docs]
-
-    Chroma.from_texts(
-        texts=texts,
-        embedding=embeddings,
-        metadatas=metadatas,
-        persist_directory=persist_dir,
+    parser = argparse.ArgumentParser(description="JRXML 向量知识库工具")
+    parser.add_argument(
+        "--download-model", action="store_true",
+        help="预下载嵌入模型到本地"
     )
-    print(f'已将 {len(docs)} 个文档索引到 Chroma，存储位置: {persist_dir}')
-
-
-if __name__ == '__main__':
-    import argparse
-    parser = argparse.ArgumentParser(description='初始化 Chroma 知识库')
-    parser.add_argument('--download-model', action='store_true', help='仅下载嵌入模型到本地')
     args = parser.parse_args()
 
     if args.download_model:
-        download_embeddings_model()
+        download_model()
     else:
-        main()
+        print("用法: python scripts/init_kb.py --download-model")
+        print()
+        print("知识库构建请在 rag/ 子项目中独立运行:")
+        print("  cd rag")
+        print("  python batch_chunker.py jrxml_source")
+        print("  python embed_chunks.py")
+        print("  python import_to_chroma.py")
+
+
+if __name__ == "__main__":
+    main()