claude-smart 0.2.43 → 0.2.44

package/plugin/vendor/reflexio/reflexio/cli/commands/setup_cmd.py

@@ -26,7 +26,11 @@ app = typer.Typer(
 )
 
 _PROVIDERS: dict[str, dict[str, str]] = {
-    "openai": {"env_var": "OPENAI_API_KEY", "model": "gpt-5-mini", "display": "OpenAI"},
+    "openai": {
+        "env_var": "OPENAI_API_KEY",
+        "model": "gpt-5.4-mini",
+        "display": "OpenAI",
+    },
     "anthropic": {
         "env_var": "ANTHROPIC_API_KEY",
         "model": "claude-sonnet-4-6",
@@ -885,9 +889,8 @@ def openai_codex_setup(
     reads from this file directly — no dependency on OpenClaw or any other
     CLI. The proxy auto-refreshes the access token when it nears expiry.
 
-    Run this once, then start the codex proxy with::
-
-        ./reflexio_ext/scripts/start_with_codex_proxy.sh
+    Run this once; the codex proxy (``codex_proxy.py``) then picks up the
+    stored tokens automatically on start.
 
     Re-run this command if your subscription tier changes or the
     refresh_token gets revoked (rare).

package/plugin/vendor/reflexio/reflexio/cli/env_loader.py

@@ -107,8 +107,8 @@ def load_reflexio_env(
 
     Args:
         package_data_module: Module containing bundled .env.example
-            (for importlib.resources). OS package uses "reflexio.data",
-            enterprise uses "reflexio_ext.data".
+            (for importlib.resources). The OS package passes "reflexio.data";
+            a downstream build may pass its own data module.
         auto_generate_keys: Env var names to auto-generate as hex tokens
             (e.g., ["JWT_SECRET_KEY"]).
 
@@ -150,8 +150,8 @@ def ensure_user_env_for_setup(
 
     Args:
         package_data_module: Module containing the bundled ``.env.example``
-            template (for ``importlib.resources``). OS uses ``reflexio.data``,
-            enterprise uses ``reflexio_ext.data``.
+            template (for ``importlib.resources``). The OS package passes
+            ``reflexio.data``; a downstream build may pass its own data module.
         auto_generate_keys: Env var names to auto-fill with random hex
             tokens when creating from the template.
 

package/plugin/vendor/reflexio/reflexio/lib/_storage_labels.py

@@ -24,8 +24,8 @@ def describe_storage(
     to print anywhere.
 
     Enterprise-only storage types (Supabase, Postgres, local directory)
-    are matched by class name so we don't take a hard import dependency
-    on ``reflexio_ext`` from the open-source package.
+    are matched by class name so the open-source package takes no hard
+    import dependency on its enterprise extension.
 
     Args:
         storage_config: The storage configuration to describe.

package/plugin/vendor/reflexio/reflexio/models/api_schema/domain/entities.py

@@ -233,7 +233,9 @@ class UserProfile(BaseModel):
     custom_features: dict | None = None
     source: str | None = None
     status: Status | None = None  # indicates the status of the profile
-    extractor_names: list[str] | None = None  # Retained provenance data column (merged on dedup); new profiles write None.
+    extractor_names: list[str] | None = (
+        None  # Retained provenance data column (merged on dedup); new profiles write None.
+    )
     expanded_terms: str | None = None
     embedding: EmbeddingVector = []
     source_span: str | None = None
@@ -554,6 +556,7 @@ class InteractionData(BaseModel):
 
 # publish user interaction request
 class PublishUserInteractionRequest(BaseModel):
+    request_id: NonEmptyStr | None = None
     user_id: NonEmptyStr
     interaction_data_list: list[InteractionData] = Field(min_length=1)
     source: str = ""
@@ -891,7 +894,9 @@ class RerunProfileGenerationRequest(BaseModel):
     start_time: datetime | None = None
     end_time: datetime | None = None
     source: str | None = None
-    extractor_names: list[str] | None = None  # Deprecated compatibility field; ignored for selection.
+    extractor_names: list[str] | None = (
+        None  # Deprecated compatibility field; ignored for selection.
+    )
 
     @model_validator(mode="after")
     def check_time_range(self) -> Self:
@@ -936,7 +941,9 @@ class ManualPlaybookGenerationRequest(BaseModel):
 
     agent_version: str = DEFAULT_AGENT_VERSION
     source: str | None = None
-    playbook_name: str | None = None  # Deprecated compatibility field; ignored for selection.
+    playbook_name: str | None = (
+        None  # Deprecated compatibility field; ignored for selection.
+    )
 
     @field_validator("agent_version")
     @classmethod
@@ -956,7 +963,9 @@ class RerunPlaybookGenerationRequest(BaseModel):
     agent_version: str = DEFAULT_AGENT_VERSION
     start_time: datetime | None = None
     end_time: datetime | None = None
-    playbook_name: str | None = None  # Deprecated compatibility field; ignored for selection.
+    playbook_name: str | None = (
+        None  # Deprecated compatibility field; ignored for selection.
+    )
     source: str | None = None
 
     @field_validator("agent_version")

package/plugin/vendor/reflexio/reflexio/models/api_schema/retriever_schema.py

@@ -56,7 +56,9 @@ class SearchUserProfileRequest(BaseModel):
     top_k: int | None = Field(default=10, gt=0)
     source: str | None = None
     custom_feature: str | None = None
-    extractor_name: str | None = None  # Deprecated compatibility field; accepted but ignored.
+    extractor_name: str | None = (
+        None  # Deprecated compatibility field; accepted but ignored.
+    )
     threshold: float | None = Field(default=0.4, ge=0.0, le=1.0)
     enable_reformulation: bool | None = False
     search_mode: SearchMode = SearchMode.HYBRID

package/plugin/vendor/reflexio/reflexio/models/api_schema/validators.py

@@ -17,6 +17,7 @@ from __future__ import annotations
 import ipaddress
 import os
 import re
+import socket
 from typing import Annotated, Any
 from urllib.parse import urlparse
 
@@ -125,6 +126,33 @@ def _is_strict_mode() -> bool:
     )
 
 
+def _is_metadata_ip(ip: ipaddress.IPv4Address | ipaddress.IPv6Address) -> bool:
+    """Return whether an IP address is a known cloud metadata endpoint."""
+    return str(ip) in METADATA_IPS
+
+
+def _is_forbidden_ip(ip: ipaddress.IPv4Address | ipaddress.IPv6Address) -> bool:
+    """Return whether an IP address is unsafe for server-side HTTP callbacks."""
+    return (
+        _is_metadata_ip(ip)
+        or ip.is_private
+        or ip.is_reserved
+        or ip.is_loopback
+        or ip.is_link_local
+        or ip.is_multicast
+        or ip.is_unspecified
+    )
+
+
+def _resolved_host_ips(host: str, port: int | None) -> set[str]:
+    """Resolve a hostname into IP address strings for SSRF validation."""
+    try:
+        addr_info = socket.getaddrinfo(host, port, type=socket.SOCK_STREAM)
+    except socket.gaierror as exc:
+        raise ValueError(f"URL hostname could not be resolved: {host}") from exc
+    return {str(entry[4][0]) for entry in addr_info}
+
+
 def _check_safe_url(v: Any) -> Any:
     """Validate URL is not targeting dangerous resources.
 
@@ -150,13 +178,11 @@ def _check_safe_url(v: Any) -> Any:
 
     try:
         ip = ipaddress.ip_address(host)
-        if str(ip) in METADATA_IPS:
+        if _is_metadata_ip(ip):
             raise ValueError(f"URL must not target cloud metadata endpoint: {host}")
 
         # In strict mode, also block private/localhost
-        if _is_strict_mode() and (
-            ip.is_private or ip.is_reserved or ip.is_loopback or ip.is_link_local
-        ):
+        if _is_strict_mode() and _is_forbidden_ip(ip):
             raise ValueError(
                 f"URL targets private/reserved IP '{host}'. "
                 f"If running locally, unset REFLEXIO_BLOCK_PRIVATE_URLS."
@@ -164,12 +190,39 @@ def _check_safe_url(v: Any) -> Any:
     except ValueError as e:
         if "must not target" in str(e) or "targets private" in str(e):
             raise
-        # Not an IP (hostname) — check localhost in strict mode
-        if _is_strict_mode() and host in ("localhost", "0.0.0.0"):  # noqa: S104
+        # Not an IP literal (hostname). Resolve and inspect each IP: cloud
+        # metadata endpoints are ALWAYS blocked (no legitimate use case),
+        # while private/reserved ranges are blocked only in strict mode.
+        strict = _is_strict_mode()
+        if strict and host in ("localhost", "0.0.0.0"):  # noqa: S104
             raise ValueError(
                 f"URL targets '{host}'. "
                 f"If running locally, unset REFLEXIO_BLOCK_PRIVATE_URLS."
             ) from None
+        try:
+            resolved_ips = _resolved_host_ips(host, parsed.port)
+        except ValueError:
+            # A resolution failure is only fatal in strict mode; otherwise we
+            # cannot confirm a metadata target and must not break local dev.
+            if strict:
+                raise
+            resolved_ips = set()
+        for resolved_ip in resolved_ips:
+            try:
+                ip = ipaddress.ip_address(resolved_ip)
+            except ValueError:
+                continue
+            if _is_metadata_ip(ip):
+                raise ValueError(
+                    f"URL hostname '{host}' resolves to cloud metadata "
+                    f"endpoint '{resolved_ip}'."
+                ) from None
+            if strict and _is_forbidden_ip(ip):
+                raise ValueError(
+                    f"URL hostname '{host}' resolves to private/reserved IP "
+                    f"'{resolved_ip}'. If running locally, unset "
+                    f"REFLEXIO_BLOCK_PRIVATE_URLS."
+                ) from None
 
     return v
 

package/plugin/vendor/reflexio/reflexio/models/config_schema.py

@@ -18,7 +18,7 @@ EMBEDDING_DIMENSIONS = 512
 
 # Default sliding window parameters for extraction
 DEFAULT_WINDOW_SIZE = 10
-DEFAULT_STRIDE_SIZE = 5
+DEFAULT_STRIDE_SIZE = 8
 
 # Deprecated aliases kept for older imports.
 DEFAULT_BATCH_SIZE = DEFAULT_WINDOW_SIZE

package/plugin/vendor/reflexio/reflexio/server/README.md

@@ -54,6 +54,8 @@ Description: FastAPI backend server that processes user interactions to generate
 
 **Directory**: `api_endpoints/`
 
+**Detailed Documentation**: See [`api_endpoints/README.md`](api_endpoints/README.md) for the `RequestContext` contract and per-file handler map.
+
 | File | Purpose |
 |------|---------|
 | `request_context.py` | RequestContext (bundles org_id, storage, configurator, prompt_manager) |
@@ -77,7 +79,7 @@ Description: FastAPI backend server that processes user interactions to generate
 - **Operations/admin**: `GET /api/get_operation_status`, `POST /api/cancel_operation`, `POST /api/admin/cache/invalidate`, `DELETE /api/delete_interaction`, `DELETE /api/delete_request`, `DELETE /api/delete_session`, `DELETE /api/delete_requests_by_ids`, `DELETE /api/delete_all_interactions`, `POST /api/clear_user_data`
 - **Human clarification/stall state**: `GET /api/pending_tool_calls`, `GET /api/pending_tool_calls/{pending_tool_call_id}`, `POST /api/pending_tool_calls/{pending_tool_call_id}/resolve`, `PATCH /api/pending_tool_calls/{pending_tool_call_id}/answer`, `POST /api/pending_tool_calls/{pending_tool_call_id}/not_applicable`, `POST /api/pending_tool_calls/{pending_tool_call_id}/cancel`, `GET /api/stall_state`, `POST /api/stall_state/notified`
 
-**Authentication Pattern**: The open-source app uses `default_get_org_id` and `DEFAULT_ORG_ID` for local/no-auth starts. Enterprise wraps `create_app()` in `reflexio_ext/server/api.py` with authenticated org resolution, login/OAuth/account/share/waitlist routers, admin checks, Sentry tracing, and usage metrics.
+**Authentication Pattern**: The open-source app uses `default_get_org_id` and `DEFAULT_ORG_ID` for local/no-auth starts. The enterprise extension wraps `create_app()` with authenticated org resolution, login/OAuth/account/share/waitlist routers, admin checks, Sentry tracing, and usage metrics.
 
 **Pattern**: Core route handlers call `Reflexio` through `get_reflexio(org_id)`; endpoint helper files should not instantiate `Reflexio` directly.
 
@@ -172,6 +174,8 @@ python -m reflexio.server.scripts.manage_invitation_codes list --show-used
 
 **Directory**: `services/`
 
+**Detailed Documentation**: See [`services/README.md`](services/README.md) for the per-directory file index across generation, evaluation, async extraction, search, and persistence.
+
 **Service Boundary**: The service layer owns LLM orchestration, extraction, evaluation, optimization, search preparation, storage access, and long-running operation state. API endpoints should validate/authenticate requests, build `RequestContext`, and delegate into `Reflexio` or focused service helpers rather than embedding business logic.
 
 **Encapsulated Components**:
@@ -611,6 +615,8 @@ All services follow BaseGenerationService:
 ## See Also
 
 - [Code Map (root README)](../README.md) -- high-level overview of all Reflexio components
+- [API Endpoints README](api_endpoints/README.md) -- RequestContext contract and handler/helper map
+- [Services README](services/README.md) -- per-directory index of the business-logic layer
 - [Prompt Bank README](prompt/prompt_bank/README.md) -- versioned prompt template system
 - [Playbook Service README](services/playbook/README.md) -- playbook extraction, aggregation, and deduplication pipeline
 - [Site Variables README](site_var/README.md) -- global configuration and feature flags

package/plugin/vendor/reflexio/reflexio/server/api.py

@@ -1,8 +1,9 @@
 import asyncio
 import inspect
 import logging
-import threading
+import os
 from collections.abc import Callable
+from concurrent.futures import ThreadPoolExecutor
 from datetime import UTC, datetime
 from typing import Any
 
@@ -21,6 +22,7 @@ from slowapi.errors import RateLimitExceeded
 from slowapi.util import get_remote_address
 from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
 from starlette.responses import Response
+from starlette.types import ASGIApp, Message, Receive, Scope, Send
 
 from reflexio.models.api_schema.braintrust_schema import (
     BraintrustStatusResponse,
@@ -194,6 +196,51 @@ SYNC_REQUEST_TIMEOUT_SECONDS = (
 )
 SUSPICIOUS_USER_AGENTS = ["bot", "crawler", "spider", "scraper", "curl", "wget"]
 ALLOWED_EMPTY_UA_PATHS = ["/health", "/"]  # Paths that allow empty user agents
+DEFAULT_MAX_BODY_BYTES = 10 * 1024 * 1024
+REGENERATE_MAX_WORKERS = 2
+_regen_executor = ThreadPoolExecutor(
+    max_workers=REGENERATE_MAX_WORKERS,
+    thread_name_prefix="reflexio-regen",
+)
+
+
+def _resolve_cors_origins() -> list[str]:
+    """Resolve browser origins allowed to make credentialed CORS requests."""
+    configured_origins = os.getenv("REFLEXIO_ALLOWED_ORIGINS", "").strip()
+    if configured_origins:
+        origins = [
+            origin.strip().rstrip("/")
+            for origin in configured_origins.split(",")
+            if origin.strip()
+        ]
+        return origins or ["http://localhost:8080"]
+
+    frontend_url = os.getenv("FRONTEND_URL", "").strip()
+    if frontend_url:
+        return [frontend_url.rstrip("/")]
+
+    return ["http://localhost:8080"]
+
+
+def _max_body_bytes_from_env() -> int:
+    raw_value = os.getenv("REFLEXIO_MAX_BODY_BYTES", str(DEFAULT_MAX_BODY_BYTES))
+    try:
+        max_bytes = int(raw_value)
+    except ValueError:
+        logger.warning(
+            "Ignoring invalid REFLEXIO_MAX_BODY_BYTES=%r; using %s",
+            raw_value,
+            DEFAULT_MAX_BODY_BYTES,
+        )
+        return DEFAULT_MAX_BODY_BYTES
+    if max_bytes <= 0:
+        logger.warning(
+            "Ignoring non-positive REFLEXIO_MAX_BODY_BYTES=%r; using %s",
+            raw_value,
+            DEFAULT_MAX_BODY_BYTES,
+        )
+        return DEFAULT_MAX_BODY_BYTES
+    return max_bytes
 
 
 def get_rate_limit_key(request: Request) -> str:
@@ -312,6 +359,82 @@ class TimeoutMiddleware(BaseHTTPMiddleware):
             )
 
 
+class _RequestBodyTooLargeError(Exception):
+    """Raised when the streamed request body exceeds the configured limit."""
+
+
+class BodySizeLimitMiddleware:
+    """Reject requests whose declared or streamed body size exceeds the limit."""
+
+    def __init__(self, app: ASGIApp) -> None:
+        self.app = app
+
+    async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
+        from starlette.responses import JSONResponse
+
+        if scope["type"] != "http":
+            await self.app(scope, receive, send)
+            return
+
+        max_body_bytes = _max_body_bytes_from_env()
+        content_length = None
+        for name, value in scope.get("headers", []):
+            if name.lower() == b"content-length":
+                content_length = value.decode("latin-1")
+                break
+
+        if content_length is not None:
+            try:
+                body_bytes = int(content_length)
+            except ValueError:
+                body_bytes = 0
+            if body_bytes > max_body_bytes:
+                await JSONResponse(
+                    status_code=status.HTTP_413_CONTENT_TOO_LARGE,
+                    content={"detail": "Request body too large"},
+                )(scope, receive, send)
+                return
+
+        consumed_bytes = 0
+
+        async def limited_receive() -> Message:
+            nonlocal consumed_bytes
+            message = await receive()
+            if message["type"] == "http.request":
+                consumed_bytes += len(message.get("body", b""))
+                if consumed_bytes > max_body_bytes:
+                    raise _RequestBodyTooLargeError
+            return message
+
+        try:
+            await self.app(scope, limited_receive, send)
+        except _RequestBodyTooLargeError:
+            await JSONResponse(
+                status_code=status.HTTP_413_CONTENT_TOO_LARGE,
+                content={"detail": "Request body too large"},
+            )(scope, receive, send)
+
+
+class SecurityHeadersMiddleware(BaseHTTPMiddleware):
+    """Attach conservative browser security headers to every response."""
+
+    async def dispatch(
+        self, request: Request, call_next: RequestResponseEndpoint
+    ) -> Response:
+        response = await call_next(request)
+        response.headers["X-Content-Type-Options"] = "nosniff"
+        response.headers["X-Frame-Options"] = "DENY"
+        response.headers["Referrer-Policy"] = "strict-origin-when-cross-origin"
+        if (
+            request.url.scheme == "https"
+            or request.headers.get("x-forwarded-proto", "").lower() == "https"
+        ):
+            response.headers["Strict-Transport-Security"] = (
+                "max-age=31536000; includeSubDomains"
+            )
+        return response
+
+
 class CorrelationIdMiddleware(BaseHTTPMiddleware):
     """Middleware that assigns a unique correlation ID to each request.
 
@@ -931,7 +1054,9 @@ def delete_user_playbooks_by_ids(
     response_model=BulkDeleteResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def delete_all_interactions(
+    request: Request,
     org_id: str = Depends(default_get_org_id),
 ) -> BulkDeleteResponse:
     """Delete all requests and their associated interactions.
@@ -950,7 +1075,9 @@ def delete_all_interactions(
     response_model=BulkDeleteResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def delete_all_profiles(
+    request: Request,
     org_id: str = Depends(default_get_org_id),
 ) -> BulkDeleteResponse:
     """Delete all profiles.
@@ -969,7 +1096,9 @@ def delete_all_profiles(
     response_model=BulkDeleteResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def delete_all_playbooks(
+    request: Request,
     org_id: str = Depends(default_get_org_id),
 ) -> BulkDeleteResponse:
     """Delete all playbooks (both user and agent).
@@ -988,7 +1117,9 @@ def delete_all_playbooks(
     response_model=BulkDeleteResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def delete_all_user_playbooks(
+    request: Request,
     org_id: str = Depends(default_get_org_id),
 ) -> BulkDeleteResponse:
     """Delete all user playbooks (user only, not agent).
@@ -1007,7 +1138,9 @@ def delete_all_user_playbooks(
     response_model=BulkDeleteResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def delete_all_agent_playbooks(
+    request: Request,
     org_id: str = Depends(default_get_org_id),
 ) -> BulkDeleteResponse:
     """Delete all agent playbooks (agent only, not user).
@@ -1026,8 +1159,10 @@ def delete_all_agent_playbooks(
     response_model=ClearUserDataResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("10/minute")
 def clear_user_data(
-    request: ClearUserDataRequest,
+    request: Request,
+    payload: ClearUserDataRequest,
     org_id: str = Depends(default_get_org_id),
 ) -> ClearUserDataResponse:
     """Delete all rows scoped to a single ``user_id``.
@@ -1046,7 +1181,7 @@ def clear_user_data(
     Returns:
         ClearUserDataResponse: Response with per-entity deletion counts
     """
-    return publisher_api.clear_user_data(org_id=org_id, request=request)
+    return publisher_api.clear_user_data(org_id=org_id, request=payload)
 
 
 @core_router.post(
@@ -1071,7 +1206,9 @@ def get_interactions(
     response_model=GetInteractionsViewResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("30/minute")
 def get_all_interactions(
+    request: Request,
     limit: int = 100,
     org_id: str = Depends(default_get_org_id),
 ) -> GetInteractionsViewResponse:
@@ -1154,7 +1291,9 @@ def get_profiles(
     response_model=GetProfilesViewResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("30/minute")
 def get_all_profiles(
+    request: Request,
     limit: int = 100,
     status_filter: str | None = None,
     org_id: str = Depends(default_get_org_id),
@@ -1230,7 +1369,9 @@ def run_playbook_aggregation(
 
 
 @core_router.post("/api/set_config")
+@limiter.limit("10/minute")
 def set_config(
+    request: Request,
     config: dict[str, Any],
     org_id: str = Depends(default_get_org_id),
 ) -> SetConfigResponse:
@@ -1257,7 +1398,9 @@ def set_config(
 
 
 @core_router.post("/api/update_config")
+@limiter.limit("10/minute")
 def update_config(
+    request: Request,
     partial: dict[str, Any],
     org_id: str = Depends(default_get_org_id),
 ) -> SetConfigResponse:
@@ -1566,8 +1709,10 @@ def update_user_profile_endpoint(
     response_model=GetDashboardStatsResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("30/minute")
 def get_dashboard_stats(
-    request: GetDashboardStatsRequest,
+    request: Request,
+    payload: GetDashboardStatsRequest,
     org_id: str = Depends(default_get_org_id),
 ) -> GetDashboardStatsResponse:
     """Get comprehensive dashboard statistics including counts and time-series data.
@@ -1583,7 +1728,7 @@ def get_dashboard_stats(
     reflexio = get_reflexio(org_id=org_id)
 
     # Get dashboard stats using Reflexio's method
-    return reflexio.get_dashboard_stats(request)
+    return reflexio.get_dashboard_stats(payload)
 
 
 @core_router.post(
@@ -1755,7 +1900,9 @@ def get_evaluation_overview(
     response_model=RegenerateStartResponse,
     response_model_exclude_none=True,
 )
+@limiter.limit("5/minute")
 def start_regenerate(
+    request: Request,
     payload: RegenerateRequest,
     org_id: str = Depends(default_get_org_id),
 ) -> RegenerateStartResponse:
@@ -1789,15 +1936,12 @@ def start_regenerate(
         )
     except RuntimeError as e:
         raise HTTPException(status_code=409, detail=str(e)) from e
-    threading.Thread(
-        target=run_regen,
-        kwargs={
-            "job": job,
-            "request_context": reflexio.request_context,
-            "llm_client": reflexio.llm_client,
-        },
-        daemon=True,
-    ).start()
+    _regen_executor.submit(
+        run_regen,
+        job=job,
+        request_context=reflexio.request_context,
+        llm_client=reflexio.llm_client,
+    )
     return RegenerateStartResponse(job_id=job.job_id, total=job.total)
 
 
@@ -2612,24 +2756,16 @@ def create_app(
     @asynccontextmanager
     async def lifespan(app: FastAPI) -> AsyncIterator[None]:  # noqa: ARG001
         validate_llm_availability()
-        bootstrap_org_id = _lifespan_org_id()
-        bootstrap_context = RequestContext(org_id=bootstrap_org_id)
-        retrieval_floor = bootstrap_context.configurator.get_config().retrieval_floor
-        if retrieval_floor.enabled:
-            from reflexio.server.llm.rerank import prewarm as _prewarm_cross_encoder
-
-            _prewarm_cross_encoder()
-        else:
-            logger.info(
-                "Cross-encoder pre-warm skipped because retrieval_floor.enabled=false"
-            )
+        from reflexio.server.llm.rerank import prewarm as _prewarm_cross_encoder
+
+        _prewarm_cross_encoder()
         # The scheduler discovers every org with resumable work each tick and
         # drives a per-org worker with org-scoped claims, so it is not limited
         # to the bootstrap org. The bootstrap org is only used to read config
         # and to seed cross-org discovery.
         scheduler = maybe_start_resume_scheduler(
             lambda org_id: RequestContext(org_id=org_id),
-            bootstrap_org_id=bootstrap_org_id,
+            bootstrap_org_id=_lifespan_org_id(),
         )
         try:
             yield
@@ -2661,14 +2797,32 @@ def create_app(
     app.add_exception_handler(RateLimitExceeded, _rate_limit_exceeded_handler)  # type: ignore[reportArgumentType]
 
     # CORS
-    origins = ["*"]
-    app.add_middleware(
-        CORSMiddleware,
-        allow_origins=origins,
-        allow_credentials=True,
-        allow_methods=["*"],
-        allow_headers=["*"],
-    )
+    # The locked-down, credentialed allowlist is an enterprise concern: only
+    # hosts that wire in auth (``require_auth=True``) restrict browser origins.
+    # OSS/local runs have no auth and bundle their own docs playground on a
+    # separate port, so they allow any origin (no credentials needed).
+    if require_auth:
+        app.add_middleware(
+            CORSMiddleware,
+            allow_origins=_resolve_cors_origins(),
+            allow_credentials=True,
+            allow_methods=["*"],
+            allow_headers=["*"],
+        )
+    else:
+        app.add_middleware(
+            CORSMiddleware,
+            allow_origins=["*"],
+            allow_credentials=False,
+            allow_methods=["*"],
+            allow_headers=["*"],
+        )
+
+    # Reject oversized requests before they reach endpoint handlers.
+    app.add_middleware(BodySizeLimitMiddleware)
+
+    # Security headers
+    app.add_middleware(SecurityHeadersMiddleware)
 
     # Timeout middleware
     app.add_middleware(TimeoutMiddleware)