fix(llm): pass LLM_EXTRA_HEADERS through ModelSettings so they reach the agent loop (#937)

strix/core/inputs.py

@@ -132,12 +132,14 @@ def make_model_settings(
     force_required_tool_choice: bool = False,
     request_timeout: float | None = None,
     prompt_cache: bool = True,
+    extra_headers: dict[str, str] | None = None,
 ) -> ModelSettings:
     model_settings = ModelSettings(
         parallel_tool_calls=False,
         retry=DEFAULT_MODEL_RETRY,
         include_usage=True,
         extra_args=request_timeout_extra_args(request_timeout),
+        extra_headers=dict(extra_headers) if extra_headers else None,
     )
     if (
         reasoning_effort is not None

strix/core/runner.py

@@ -250,6 +250,7 @@ async def run_strix_scan(
             force_required_tool_choice=settings.llm.force_required_tool_choice,
             request_timeout=settings.llm.timeout,
             prompt_cache=settings.llm.prompt_cache,
+            extra_headers=settings.llm.extra_headers,
         )
         run_config = RunConfig(
             model=resolved_model,