fix(types): widen untyped request dicts for the allowed_tools refusal on lts/1.108.x

The line types the hook loop's kwargs and optional_params as plain
dicts, so passing them to the picked refusal helpers added unknown
argument errors that main does not have. They are now cast to
Mapping[str, object] at the call sites, and the loop settings reader
takes a Mapping

The new cast import in the agentic loop would grow the banned-API
count, so the duplicate function-level cast import in
llm_http_handler.py goes, since the module already imports cast

litellm/litellm_core_utils/chat_completion_agentic_loop.py

@@ -4,7 +4,7 @@ import json
 from collections.abc import Iterable, Mapping
 from itertools import chain
 from types import MappingProxyType
-from typing import Final
+from typing import Final, cast
 
 from pydantic import TypeAdapter, ValidationError
 
@@ -85,7 +85,7 @@ def _coerce_int(value: object, default: int) -> int:
     return int(value) if isinstance(value, (int, str)) else default
 
 
-def _agentic_loop_settings(kwargs: dict[str, object]) -> tuple[int, int, list[str]]:
+def _agentic_loop_settings(kwargs: Mapping[str, object]) -> tuple[int, int, list[str]]:
     depth: Final = _coerce_int(kwargs.get("_agentic_loop_depth"), 0)
     configured: Final = validated_max_agentic_loops(
         kwargs.get("max_agentic_loops"), field="litellm_params.max_agentic_loops"
@@ -334,7 +334,10 @@ async def maybe_run_chat_completion_agentic_loop(
     import litellm
 
     callbacks: Final = litellm.callbacks + (getattr(logging_obj, "dynamic_success_callbacks", None) or [])
-    request_kwargs: Final = with_caller_tool_choice(kwargs, optional_params)
+    requested_optional_params: Final = cast(  # cast-ok: optional_params is a str-keyed dict
+        "Mapping[str, object]", optional_params
+    )
+    request_kwargs: Final = with_caller_tool_choice(kwargs, requested_optional_params)
     depth, max_loops, fingerprints = _agentic_loop_settings(request_kwargs)
     tools: Final = optional_params.get("tools", [])
 

litellm/llms/custom_httpx/llm_http_handler.py

@@ -5965,8 +5965,6 @@ class BaseLLMHTTPHandler:
             else False
         )
         if websearch_converted_stream and isinstance(response, dict):
-            from typing import cast
-
             from litellm._logging import verbose_logger
             from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
                 FakeAnthropicMessagesStreamIterator,
@@ -6209,6 +6207,7 @@ class BaseLLMHTTPHandler:
         callbacks: Final = litellm.callbacks + (logging_obj.dynamic_success_callbacks or [])
         tools: Final = optional_params.get("tools", [])
         depth, max_loops, fingerprints = self._get_agentic_loop_settings(kwargs=kwargs)
+        request_kwargs: Final = cast("Mapping[str, object]", kwargs)  # cast-ok: kwargs is a str-keyed dict
 
         for callback in callbacks:
             if not isinstance(callback, CustomLogger):
@@ -6243,7 +6242,7 @@ class BaseLLMHTTPHandler:
 
             # Safety guards must run OUTSIDE the callback try/except — they are
             # bounded-loop / cycle-break rails that must propagate to the caller.
-            refuse_a_call_the_choice_excludes(response, kwargs, callbacks, model)
+            refuse_a_call_the_choice_excludes(response, request_kwargs, callbacks, model)
             fingerprint = self._check_agentic_loop_safety(
                 tool_calls=tool_calls,
                 fingerprints=fingerprints,