feat(memoir): 回忆录分段两阶段管线（Phase1 分类 / Phase2 叙事）与配置、测试

api/app/agents/memoir/classification_agent.py

@@ -14,8 +14,11 @@ from dataclasses import dataclass
 from typing import Any
 
 from app.agents.memoir.prompts import get_chapter_classification_json_prompt
-from app.agents.stage_constants import CHAPTER_CATEGORIES
-from app.agents.stage_constants import STAGE_TO_DEFAULT_CATEGORY
+from app.agents.stage_constants import (
+    CHAPTER_CATEGORIES,
+    STAGE_KEYWORD_WEIGHTS,
+    STAGE_TO_DEFAULT_CATEGORY,
+)
 from app.core.json_utils import extract_json_payload
 from app.core.langchain_llm import invoke_json_object
 from app.core.logging import get_logger
@@ -40,21 +43,12 @@ _SHORT_HUKOU_STYLE = re.compile(
     re.UNICODE,
 )
 
-# 5-stage 关键词（用于 LLM 失败时的兜底）；注意勿含易与「仅年份句」共现的泛词，以免误推类别
-STAGE_KEYWORDS = {
-    "childhood": ["童年", "小时候", "家乡", "小镇"],
-    "education": ["上学", "学校", "老师", "同学", "教育", "大学"],
-    "career": ["工作", "职业", "事业", "公司", "同事", "创业"],
-    "family": ["伴侣", "孩子", "家庭", "家人", "结婚", "父母"],
-    "belief": ["信念", "价值观", "座右铭", "坚持", "原则"],
-}
-
 
 def _detect_stage(text: str, fallback_stage: str) -> str:
-    """根据关键词检测消息所属的 5-stage 阶段"""
+    """根据关键词检测消息所属的 5-stage 阶段（与 stage_constants.STAGE_KEYWORD_WEIGHTS 同源；匹配方式为子串，非加权）。"""
     message = (text or "").lower()
-    for stage, keywords in STAGE_KEYWORDS.items():
-        if any(word in message for word in keywords):
+    for stage, pairs in STAGE_KEYWORD_WEIGHTS.items():
+        if any(word in message for word, _w in pairs):
             return stage
     return fallback_stage