Safety Check Utils

Shared helper for NeMo Guardrails safety check actions.

execute_nemo_safety_check(llm_task_manager, llm, task, prompt_context, log_label) async

Render a NeMo safety prompt, call the LLM, and parse the result.

Encapsulates the full render → call → parse pipeline shared by both input and output safety check actions. Returns [True] (fail open) on any exception so the Colang flow can always read $response[0] without crashing.

Parameters:

Name	Type	Description	Default
llm_task_manager	LLMTaskManager	The LLM task manager.	required
llm	Any | None	NeMo 0.22+ LLMModel instance used for the check. The type is Any because NeMo 0.22 decoupled from LangChain and now passes a protocol implementation instead of a BaseLLM/BaseLanguageModel instance.	required
task	Task	The NeMo task type (e.g. SELF_CHECK_INPUT).	required
prompt_context	dict[str, Any]	Template variables passed to render_task_prompt.	required
log_label	str	Label used in log messages (e.g. "Input" or "Output").	required

Returns:

Type	Description
list[bool | str]	list[bool | str]: [True] if safe, [False, reason] if unsafe, or [True] as a fail-open sentinel on any runtime error.