fix(rag): keep practice headings with details

ai_knowledge_base_v2/apps/backend/tests/test_knowledge_agent.py

@@ -21,6 +21,7 @@ from app.models.knowledge import (
 )
 from app.models.chat import ChatMessage
 from app.services.knowledge_agent_service import Candidate, KnowledgeAgentService
+from app.services.knowledge_pipeline_service import parse_sections
 
 
 def _database() -> Session:
@@ -192,6 +193,10 @@ def test_homework_overview_expands_practice_terms_and_section_limit():
     assert KnowledgeAgentService._title_intent_boost("十七、练习一：风铃式静心", terms) == 20.0
     assert KnowledgeAgentService._title_intent_boost("完整练习的方向", terms) == 3.0
     assert KnowledgeAgentService._title_intent_boost("课程定位", terms) == 0.0
+    assert KnowledgeAgentService._is_practice_overview("原生里的作业内容都有什么") is True
+    assert KnowledgeAgentService._is_practice_overview(
+        "关于“心光里都有什么功课”的追问：金色光练习具体内容是什么"
+    ) is False
 
 
 def test_homework_overview_keeps_all_numbered_practices_before_selection():
@@ -264,3 +269,110 @@ def test_complete_section_includes_child_headings_but_stops_at_next_peer():
 
         assert "鼻吸鼻呼" in content
         assert "下一项作业" not in content
+
+
+def test_same_level_practice_headings_are_read_as_one_semantic_unit():
+    with _database() as db:
+        knowledge = _add_published_knowledge(db, knowledge_id=1, name="心光课程")
+        version_id = knowledge.current_version_id
+        parent = db.scalar(select(KnowledgeSection).where(KnowledgeSection.version_id == version_id))
+        assert parent is not None
+        parent.title = "二十九、练习一：金色光欧姆静心"
+        parent.content = "# 二十九、练习一：金色光欧姆静心"
+        parent.sort_order = 1
+        parent_chunk = db.scalar(select(KnowledgeChunk).where(KnowledgeChunk.section_id == parent.id))
+        assert parent_chunk is not None
+        parent_chunk.title = parent.title
+        parent_chunk.content = parent.content
+        parent_chunk.normalized_text = "金色光 欧姆 静心 练习"
+        detail = KnowledgeSection(
+            knowledge_id=knowledge.id,
+            version_id=version_id,
+            section_key="S0002",
+            title="基础原则",
+            content="# 基础原则\n练习时保持自然呼吸，不追求必须看见金色光。",
+            source_start=10,
+            source_end=50,
+            sort_order=2,
+            content_hash="detail",
+        )
+        db.add_all([
+            parent,
+            parent_chunk,
+            detail,
+            KnowledgeSection(
+                knowledge_id=knowledge.id,
+                version_id=version_id,
+                section_key="S0003",
+                title="三十、金色光练习的现实边界",
+                content="# 三十、金色光练习的现实边界\n不能替代医疗诊断。",
+                source_start=51,
+                source_end=80,
+                sort_order=3,
+                content_hash="boundary",
+            ),
+            KnowledgeSection(
+                knowledge_id=knowledge.id,
+                version_id=version_id,
+                section_key="S0004",
+                title="三十一、练习二：纠缠之心静心",
+                content="# 三十一、练习二：纠缠之心静心\n下一项练习内容。",
+                source_start=81,
+                source_end=120,
+                sort_order=4,
+                content_hash="next",
+            ),
+        ])
+        db.flush()
+        db.add(
+            KnowledgeChunk(
+                knowledge_id=knowledge.id,
+                version_id=version_id,
+                section_id=detail.id,
+                title=detail.title,
+                content=detail.content,
+                normalized_text="金色光 基础原则 自然呼吸 画面",
+                keywords='["金色光","自然呼吸"]',
+                synonyms="[]",
+                source_start=detail.source_start,
+                source_end=detail.source_end,
+                sort_order=1,
+                content_hash="detail-chunk",
+            )
+        )
+        db.commit()
+
+        content = KnowledgeAgentService._read_complete_section(db, detail)
+        result = asyncio.run(
+            KnowledgeAgentService.build_result(db, question="金色光欧姆静心具体怎么练习？")
+        )
+
+        assert "金色光欧姆静心" in content
+        assert "保持自然呼吸" in content
+        assert "不能替代医疗诊断" in content
+        assert "下一项练习内容" not in content
+        assert result.chunks
+        assert "保持自然呼吸" in result.chunks[0].content
+
+
+def test_parser_merges_practice_title_and_same_level_detail_headings():
+    source = """# 二十九、练习一：金色光欧姆静心
+# 练习定位
+帮助学员稳定注意力。
+# 基础原则
+保持自然呼吸，不追求画面。
+# 三十、金色光练习的现实边界
+不能替代医疗诊断。
+# 三十一、练习二：纠缠之心静心
+这是下一项练习。
+"""
+
+    sections = parse_sections(source)
+
+    assert [section.title for section in sections] == [
+        "二十九、练习一：金色光欧姆静心",
+        "三十一、练习二：纠缠之心静心",
+    ]
+    assert "基础原则" in sections[0].content
+    assert "现实边界" in sections[0].content
+    assert "下一项练习" not in sections[0].content