fix(vector-sync): address round-2 review — one-shot backstop, helper, caps

- scanner: gate the consent backstop with a per-(user,doc_type) one-shot
  marker so a standing admin-disable doesn't re-enqueue idempotent deletes
  every scan tick; the marker clears when the type is re-enabled. Derive
  _TEXT_BACKSTOP_DOC_TYPES from INDEXED_DOC_TYPES so new indexed types are
  covered automatically
- semantic: extract _consent_narrowed_doc_types so the search-side narrowing
  is unit-testable; add tests for restrict/intersect/disjoint/empty
- purge route: cap doc_types length (abuse guard) -> 400
- tests: one-shot + re-enable backstop, too-many-doc_types 400

Deferred (noted on PR): per-document allowed_doc_types call is cache-hot;
purge "last error wins" — both logged. SonarCloud broad-except hotspots are
deliberate (noqa BLE001), reviewable in the UI.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

nextcloud_mcp_server/api/vector_sync.py

@@ -32,6 +32,11 @@ from ..http import nextcloud_httpx_client
 
 logger = logging.getLogger(__name__)
 
+# Upper bound on doc_types per purge request. There are only a handful of real
+# indexed types; this caps a hostile/buggy caller's fan-out of count+delete
+# calls without constraining legitimate use.
+_MAX_PURGE_DOC_TYPES = 64
+
 
 async def purge_doc_types_route(request: Request) -> JSONResponse:
     """POST /api/v1/vector-sync/purge — delete indexed vectors by doc type.
@@ -83,6 +88,16 @@ async def purge_doc_types_route(request: Request) -> JSONResponse:
     doc_types = [d for d in raw if d]
     if not doc_types:
         return JSONResponse({"purged": {}})
+    # Bound the batch: there are only a handful of real indexed types, so a huge
+    # list is abuse — cap it rather than fan out unbounded count+delete calls.
+    if len(doc_types) > _MAX_PURGE_DOC_TYPES:
+        return JSONResponse(
+            {
+                "error": "Bad request",
+                "message": f"doc_types exceeds the maximum of {_MAX_PURGE_DOC_TYPES}",
+            },
+            status_code=400,
+        )
 
     try:
         username, app_password = await get_basic_auth_for_user(user_id)

nextcloud_mcp_server/server/semantic.py

@@ -57,6 +57,23 @@ logger = logging.getLogger(__name__)
 _USAGE_METADATA_MAX_DOC_TYPES = 16
 
 
+def _consent_narrowed_doc_types(
+    doc_types: list[str] | None, allowed: frozenset[str]
+) -> list[str]:
+    """Apply the admin allow-set to a requested ``doc_types`` filter.
+
+    Caller has already established ``allowed is not None`` (a concrete allow-set;
+    ``None`` means "no restriction" and is handled by skipping this call). When
+    no explicit ``doc_types`` are requested, restrict to the full allow-set;
+    otherwise intersect (preserving the caller's order). An empty result means
+    nothing the caller asked for is admin-approved — the caller short-circuits
+    to an empty response rather than falling through to an all-types search.
+    """
+    if doc_types is None:
+        return sorted(allowed)
+    return [dt for dt in doc_types if dt in allowed]
+
+
 async def record_search_usage(
     *,
     enabled: bool,
@@ -309,10 +326,7 @@ def configure_semantic_tools(mcp: FastMCP):
         # means the admin disabled every source.
         allowed = await allowed_doc_types(client, username)
         if allowed is not None:
-            if doc_types is None:
-                doc_types = sorted(allowed)
-            else:
-                doc_types = [dt for dt in doc_types if dt in allowed]
+            doc_types = _consent_narrowed_doc_types(doc_types, allowed)
             if not doc_types:
                 logger.info(
                     "Semantic search short-circuited for user %s: no requested "

nextcloud_mcp_server/vector/scanner.py

@@ -282,9 +282,16 @@ def _app_enabled(app_id: str, enabled_apps: set[str] | None) -> bool:
 
 # Text doc types whose deletion-tracking lives *inside* their scan_* function,
 # so skipping that function (when admin-disabled) leaves indexed points with no
-# grace-period backstop. ``file`` is intentionally excluded: its scan path
-# empties discovery and lets the existing reconcile loop purge on disable.
-_TEXT_BACKSTOP_DOC_TYPES: tuple[str, ...] = ("note", "news_item", "deck_card")
+# grace-period backstop. Derived from INDEXED_DOC_TYPES so a newly-indexed type
+# automatically gets the backstop. ``file`` is excluded: its scan path empties
+# discovery and lets the existing reconcile loop purge on disable.
+_TEXT_BACKSTOP_DOC_TYPES: tuple[str, ...] = tuple(sorted(INDEXED_DOC_TYPES - {"file"}))
+
+# Per-process record of (user_id, doc_type) whose consent backstop deletes have
+# already been enqueued, so a *standing* admin-disable doesn't re-flood the
+# processor with idempotent deletes on every scan tick. An entry is cleared once
+# the type is allowed again, so a later re-disable re-triggers the backstop.
+_consent_backstop_done: set[tuple[str, str]] = set()
 
 
 async def _enqueue_deletes_for_disabled_types(
@@ -303,7 +310,18 @@ async def _enqueue_deletes_for_disabled_types(
     if allowed is None:
         return 0
 
-    disabled = [dt for dt in _TEXT_BACKSTOP_DOC_TYPES if dt not in allowed]
+    # Re-enabled types: clear their one-shot marker so a later re-disable
+    # re-triggers the backstop.
+    for doc_type in _TEXT_BACKSTOP_DOC_TYPES:
+        if doc_type in allowed:
+            _consent_backstop_done.discard((user_id, doc_type))
+
+    # Disabled types not yet backstopped this episode.
+    disabled = [
+        dt
+        for dt in _TEXT_BACKSTOP_DOC_TYPES
+        if dt not in allowed and (user_id, dt) not in _consent_backstop_done
+    ]
     if not disabled:
         return 0
 
@@ -346,6 +364,9 @@ async def _enqueue_deletes_for_disabled_types(
                 )
             )
             queued += 1
+        # Mark this (user, doc_type) backstopped for the current disable episode
+        # so subsequent scans don't re-enqueue the same idempotent deletes.
+        _consent_backstop_done.add((user_id, doc_type))
     return queued