Assistant: stop stale clients from resurrecting cleared threads

Root cause of the "chunks" clear bug: conversation history is
server-owned, but MAF seeds a thread from whatever the client sends
when no snapshot exists. Any stale tab (open across a clear or from an
older session) still holds the old conversation in memory, and its next
question re-persisted the whole thing — clears appeared to peel off one
layer at a time.

The agent middleware now drops client-sent history for threads with no
stored snapshot, keeping only the last user turn (fresh conversations
and empty-messages hydration requests pass through untouched). Verified
live: simulated a cleared-thread stale tab, asked from it, and only the
new turn was persisted.

Also: clearThread surfaces DELETE failures as a toast instead of
silently reloading, and conftest sets a dummy OPENAI_API_KEY so tests
can import app.main.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/app/main.py

@@ -65,6 +65,40 @@ def _pair_orphan_tool_calls(messages: list) -> list:
     return out
 
 
+def _drop_stale_client_history(session, thread_id, messages: list) -> list:
+    """Conversation history is server-owned (the AG-UI snapshot store).
+
+    A request that carries assistant/tool history for a thread with NO
+    stored snapshot comes from a stale client — typically a second tab that
+    was open when 'Nueva conversación' cleared the thread. Trusting it
+    would re-persist the cleared conversation ('chunks' resurrecting after
+    every clear). Keep only the last user turn; MAF starts the thread
+    fresh from there.
+    """
+    if not thread_id or not messages:
+        return messages
+    has_history = any(m.get("role") in ("assistant", "tool") for m in messages)
+    if not has_history:
+        return messages
+    from sqlmodel import select
+
+    from app.models.models import ChatThreadSnapshot
+
+    row = session.exec(
+        select(ChatThreadSnapshot.id).where(
+            ChatThreadSnapshot.thread_id == str(thread_id)
+        )
+    ).first()
+    if row is not None:
+        return messages  # thread exists; history is legitimate
+    last_user = None
+    for i in range(len(messages) - 1, -1, -1):
+        if messages[i].get("role") == "user":
+            last_user = i
+            break
+    return messages[last_user:] if last_user is not None else []
+
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     run_alembic_upgrade()
@@ -121,22 +155,27 @@ async def agent_auth_and_session(request: Request, call_next):
     except JWTError:
         return Response(status_code=401, content="Invalid token")
 
+    session_gen = get_session()
+    session = next(session_gen)
+    token_var = set_session(session)
+
     # Repair orphan tool_calls before the MAF agent sees the message history.
     if request.method == "POST" and "application/json" in request.headers.get("content-type", ""):
         raw = await request.body()
         try:
             body = json.loads(raw)
             if isinstance(body.get("messages"), list):
+                body["messages"] = _drop_stale_client_history(
+                    session,
+                    body.get("threadId") or body.get("thread_id"),
+                    body["messages"],
+                )
                 body["messages"] = _pair_orphan_tool_calls(body["messages"])
                 raw = json.dumps(body).encode()
         except Exception:
             pass
         # Starlette caches the body; replace it so call_next sees the fixed bytes.
         request._body = raw  # type: ignore[attr-defined]
-
-    session_gen = get_session()
-    session = next(session_gen)
-    token_var = set_session(session)
     # Browser's IANA timezone, forwarded by the BFF — lets tools resolve
     # "today" wherever the user is (falls back to Costa Rica).
     tz_token = set_client_timezone(request.headers.get("x-client-timezone"))