WIP: memory system improvements (in progress)

Interview/chat prompt layers, reply planner, style profiles, memory
injection, interview meta store, and related tests. Work not finished.

Made-with: Cursor

api/app/agents/chat/interview_state_hints.py

@@ -16,6 +16,8 @@ from app.agents.state_schema import KnownFact, MemoirStateSchema, PersonaThread
 _QUESTION_SPLIT_RE = re.compile(r"[?？]+")
 _SENTENCE_SPLIT_RE = re.compile(r"(?<=[。！？!?])")
 _PUNCT_RE = re.compile(r"[\s，。！？；：、“”‘’（）()《》【】\\[\\],.!?:;\"'`~·…-]+")
+# 「我演罗密欧」等扮演亲历，但排除「我演示…」类口癖
+_AUTOBIO_IYAN_NOT_DEMO_RE = re.compile(r"我演(?!示)")
 
 _TRAIT_HINTS: tuple[tuple[str, tuple[str, ...]], ...] = (
     ("执着坚持", ("坚持", "执着", "咬牙", "熬过", "顶住", "训练", "反复")),
@@ -337,6 +339,91 @@ def segments_are_only_duplicate_guard_fallback(segments: Iterable[str]) -> bool:
     return len(parts) == 1 and parts[0] == DUPLICATE_QUESTION_GUARD_FALLBACK_ZH
 
 
+# 助手可见回复中，明显声称「我本人有过某种人生经历」的高置信子串（偏保守、宁可漏网不误伤泛化共情）。
+_AUTOBIOGRAPHICAL_MARKERS_ZH: tuple[str, ...] = (
+    "我小时候",
+    "我小学",
+    "我中学",
+    "我初中",
+    "我高中",
+    "我大学",
+    "我上学那",
+    "我念书",
+    "我读书那",
+    "我暗恋",
+    "我当时暗恋",
+    "我爸妈",
+    "我父亲",
+    "我母亲",
+    "我爹",
+    "我妈",
+    "我爷爷",
+    "我奶奶",
+    "我外公",
+    "我外婆",
+    "我前任",
+    "我老公",
+    "我老婆",
+    "我丈夫",
+    "我妻子",
+    "我男友",
+    "我女友",
+    "我对象",
+    "我儿子",
+    "我女儿",
+    "我孩子",
+    "我以前也",
+    "我当时也",
+    "我那时候也",
+    "我也经历过",
+    "我也有过",
+    "我也演过",
+    "我也上台",
+    "我演过",
+    "我饰演",
+    "我演出",
+    "我演的是",
+    "我演的",
+    "我扮演",
+    "感觉我熟",
+    "这我熟",
+)
+
+AUTOBIOGRAPHICAL_BOUNDARY_FALLBACK_ZH = (
+    "你刚说的这段很有画面，我想多听你讲讲那时候你心里是什么感觉。"
+)
+
+
+def _segment_has_autobiographical_claim_zh(text: str) -> bool:
+    s = (text or "").strip()
+    if not s:
+        return False
+    if _AUTOBIO_IYAN_NOT_DEMO_RE.search(s):
+        return True
+    return any(m and m in s for m in _AUTOBIOGRAPHICAL_MARKERS_ZH)
+
+
+def apply_autobiographical_boundary_guard(
+    segments: Iterable[str],
+) -> tuple[list[str], bool]:
+    """将明显带有「助手自传式经历」的段落替换为中性承接，避免身份越界。"""
+    cleaned: list[str] = []
+    touched = False
+    for seg in segments:
+        text = str(seg or "").strip()
+        if not text:
+            continue
+        if _segment_has_autobiographical_claim_zh(text):
+            cleaned.append(AUTOBIOGRAPHICAL_BOUNDARY_FALLBACK_ZH)
+            touched = True
+        else:
+            cleaned.append(text)
+    if not cleaned:
+        cleaned = [AUTOBIOGRAPHICAL_BOUNDARY_FALLBACK_ZH]
+        touched = True
+    return cleaned, touched
+
+
 def stage_slot_hint_lines(stage: str) -> list[str]:
     keys = STAGE_SLOT_KEYS.get(stage, ())
     stage_zh = STAGE_DISPLAY_ZH.get(stage, stage)