refactor(rag): 梳理文档入库与检索服务结构

tests/test_agent_core.py

@@ -1,5 +1,5 @@
 from agent_core.orchestrator import build_messages, run_agent
-from agent_core.rag.ingest import ingest_document
+from agent_core.rag.ingest import _split_text, ingest_document
 from agent_core.rag.retriever import retrieve
 
 
@@ -221,3 +221,30 @@ def test_run_agent_uses_retrieved_document_chunks(tmp_path):
 
     assert result.references[0]["source"] == "sop.md"
     assert "隔离现场" in result.references[0]["content"]
+
+
+def test_rag_split_text_keeps_overlap_and_non_empty_chunks():
+    chunks = _split_text("A" * 20, chunk_size=8, overlap=3)
+
+    assert chunks == ["AAAAAAAA", "AAAAAAAA", "AAAAAAAA", "AAAAA"]
+
+
+def test_retrieve_returns_empty_when_query_has_no_overlap(tmp_path):
+    store_path = tmp_path / "rag_store.json"
+    ingest_document(
+        scenario_id="knowledge_qa",
+        source_file="rules.md",
+        text="这里描述的是报销流程和审批链。",
+        collection="knowledge_qa",
+        store_path=store_path,
+    )
+
+    chunks = retrieve(
+        scenario_id="knowledge_qa",
+        query="设备点检",
+        collection="knowledge_qa",
+        top_k=3,
+        store_path=store_path,
+    )
+
+    assert chunks == []