Add read-only production runtime audit

apps/ai-service/routers/rag.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import uuid
 from typing import Annotated, Any, Literal, Protocol
 
-from fastapi import APIRouter, Depends, HTTPException, Request
+from fastapi import APIRouter, Depends, HTTPException, Query, Request
 from pydantic import BaseModel, Field
 
 from rag.answer import DISCLAIMER, GroundedAnswerService
@@ -36,6 +36,7 @@ class RagQueryRequest(BaseModel):
     # (follow-up inheritance, clarify, smalltalk). Absent → single-turn, exactly
     # as before, so existing callers are unchanged.
     conversation_id: str | None = Field(default=None, max_length=128)
+    response_mode: Literal["ai", "monograph"] = "ai"
 
 
 class CitationResponse(BaseModel):
@@ -198,7 +199,7 @@ _HISTORY_LIMIT = 50
 
 @router.get("/history", response_model=HistoryResponse)
 def list_history(
-    conversation_id: str,
+    conversation_id: Annotated[str, Query(max_length=128)],
     traces: Annotated[TraceWriter, Depends(_trace_writer)],
 ) -> HistoryResponse:
     """Feature-List #25: past queries for one session, most recent first, so
@@ -430,7 +431,11 @@ def query_rag(
         # then routes to the safety-verified retrieval + grounded-answer
         # engine. Replaces the old fuzzy resolver + keyword section router +
         # manual follow-up inheritance for both single- and multi-turn.
-        reply = agent.handle(payload.query, payload.conversation_id)
+        reply = agent.handle(
+            payload.query,
+            payload.conversation_id,
+            response_mode=payload.response_mode,
+        )
         decision = reply.decision
         reason = reply.reason
         answer = reply.clarification if reply.clarification is not None else reply.answer