세션 계약과 메모리 경계 보강

apps/api/app/routes/sessions.py

@@ -121,7 +121,6 @@ _RECALL_CACHE: dict[str, memory.RecallContext] = {}
 # 세션별 KB 증상 행동단서(회기 1회 산출·캐시). 빈 list 캐시 = 회기 내 재시도 안 함(안정성).
 _KB_CUES_CACHE: dict[str, list[str]] = {}
 _RAG_WARM_SEMAPHORE = asyncio.Semaphore(1)
-_LEARNER_VISIBLE_AI_ROLE = "counselor"
 
 # ────────────────────────────────────────────────────────────────────────────
 # RAG 배선 헬퍼 — 내담자(CLIENT) 뷰. 임베더/KB/DB 풀 미가용 시 빈 값으로 graceful
@@ -244,12 +243,6 @@ async def _ensure_kb_cues(session_id: str, card) -> list[str]:
     return cues
 
 
-async def _load_prev_case_summary(case_id: str) -> Optional[dict]:
-    """직전 회기 요약(case 스코프) → build_recall_context 입력. 미존재/미가용 시 None."""
-    case_memory = await _load_case_memory(case_id)
-    return case_memory.get("prev_summary")
-
-
 async def _load_case_memory(case_id: str) -> dict:
     """case-level 큰그림 + 직전 요약 + client-visible pinned fact를 한 번에 읽는다."""
     empty = {"case_digest": None, "prev_summary": None, "pinned_facts": []}
@@ -387,6 +380,30 @@ async def ensure_recall_context(sess: InProcSession) -> memory.RecallContext:
     return recall
 
 
+async def _prepare_turn_context(
+    *,
+    session_id: str,
+    sess: InProcSession,
+    learner_text: str,
+) -> orchestrator.TurnContext:
+    recall = await ensure_recall_context(sess)
+    kb_cues = _KB_CUES_CACHE.get(session_id) or []  # 비차단: warm 전이면 빈 단서(graceful)
+    ctx = orchestrator.prepare_turn(
+        session_id=session_id,
+        case_id=sess.case_id,
+        card=sess.persona,
+        state=sess.state,
+        learner_text=learner_text,
+        recall_summary=recall.recall_summary,
+        pinned_facts=recall.pinned_facts,
+        recent_turns=sess.recent_turns(visible_to="client"),
+        kb_behavior_cues=kb_cues,
+        theory_mode=sess.theory_mode,
+    )
+    assert ctx.state_after is not None
+    return ctx
+
+
 async def _warm_rag_caches(session_id: str, case_id: str, card) -> None:
     """RAG 회상·KB 행동단서를 **백그라운드**로 산출해 캐시한다(요청 경로 비차단).
 
@@ -1032,22 +1049,11 @@ async def submit_turn(
     """Submit one trainee utterance and return the generated client reply."""
     principal = _ensure_learner(principal)
     sess = await _load_session_or_404(session_id, principal)
-    recall = await ensure_recall_context(sess)
-    kb_cues = _KB_CUES_CACHE.get(session_id) or []  # 비차단: warm 전이면 빈 단서(graceful)
-
-    ctx = orchestrator.prepare_turn(
+    ctx = await _prepare_turn_context(
         session_id=session_id,
-        case_id=sess.case_id,
-        card=sess.persona,
-        state=sess.state,
         learner_text=body.text,
-        recall_summary=recall.recall_summary,
-        pinned_facts=recall.pinned_facts,
-        recent_turns=sess.recent_turns(visible_to="client"),
-        kb_behavior_cues=kb_cues,
-        theory_mode=sess.theory_mode,
+        sess=sess,
     )
-    assert ctx.state_after is not None
 
     try:
         result = await orchestrator.run_turn_generate(
@@ -1159,22 +1165,11 @@ async def stream_turn(
     """Stream a generated client reply for one trainee utterance."""
     principal = _ensure_learner(principal)
     sess = await _load_session_or_404(session_id, principal)
-    recall = await ensure_recall_context(sess)
-    kb_cues = _KB_CUES_CACHE.get(session_id) or []  # 비차단: warm 전이면 빈 단서(graceful)
-
-    ctx = orchestrator.prepare_turn(
+    ctx = await _prepare_turn_context(
         session_id=session_id,
-        case_id=sess.case_id,
-        card=sess.persona,
-        state=sess.state,
         learner_text=body.text,
-        recall_summary=recall.recall_summary,
-        pinned_facts=recall.pinned_facts,
-        recent_turns=sess.recent_turns(visible_to="client"),
-        kb_behavior_cues=kb_cues,
-        theory_mode=sess.theory_mode,
+        sess=sess,
     )
-    assert ctx.state_after is not None
 
     async def event_generator():
         last_beat = asyncio.get_running_loop().time()
@@ -1229,7 +1224,7 @@ async def end_session(
         session_id=session_id,
         case_id=sess.case_id,
         session_no=sess.session_no,
-        masked_turns=sess.masked_turns(),
+        masked_turns=sess.masked_turns(visible_to="client"),
         prev_rapport_credit=sess.prev_rapport_credit,
         open_threads=recall.open_threads,
     )