apply review suggestions

Author: Pouyanpi

nemoguardrails/actions/llm/utils.py

@@ -396,23 +396,28 @@ def _extract_content(response: LLMResponse) -> str:
     return response.content
 
 
-def warn_if_truncated(response: LLMResponse, task: str) -> None:
-    """Emit a warning if the LLM produced no visible content because it hit the max_tokens budget.
+def warn_if_truncated(response: LLMResponse, task: str) -> bool:
+    """Return True and emit a warning if the LLM produced no visible content because it hit the max_tokens budget.
 
     Reasoning models (OpenAI o-series, gpt-5, DeepSeek-R1, Gemini 2.5, Qwen QwQ, etc.)
     spend output tokens on internal reasoning before emitting visible text. A small
     max_tokens budget can be fully consumed by the reasoning phase, leaving empty
     content and finish_reason="length". The call succeeds silently and callers that
-    only inspect response.content see nothing.
+    only inspect response.content see nothing. Callers whose downstream parser
+    does not fail safely on empty input (e.g. self_check_facts, whose parser
+    inverts the result) should use the return value to take an explicit
+    fail-safe branch.
     """
-    if not response.content and response.finish_reason == "length":
+    truncated = not response.content and response.finish_reason == "length"
+    if truncated:
         logger.warning(
             "Task %s: LLM returned empty content with finish_reason='length'. "
             "The max_tokens budget was likely consumed before any visible output. "
             "If using a reasoning model (o1/o3/o4-mini, gpt-5, deepseek-r1, "
             "gemini-2.5, etc.), increase the prompt's max_tokens in your config.",
             task,
         )
+    return truncated
 
 
 def get_colang_history(

nemoguardrails/library/self_check/facts/actions.py

@@ -76,7 +76,11 @@ async def self_check_facts(
         stop=stop,
         llm_params={"temperature": config.lowest_temperature, "max_tokens": max_tokens},
     )
-    warn_if_truncated(llm_response, task.value)
+    if warn_if_truncated(llm_response, task.value):
+        # is_content_safe returns [False] on empty input, which this action
+        # inverts to "facts are correct". Fail safe instead: treat truncated
+        # content as a failed fact-check so the output gets blocked.
+        return 0.0
     response = llm_response.content
 
     if llm_task_manager.has_output_parser(task):

tests/integrations/langchain/test_actions_llm_utils.py

@@ -649,21 +649,33 @@ def test_warns_on_empty_content_with_length_finish(self, caplog):
 
         response = LLMResponse(content="", finish_reason="length")
         with caplog.at_level("WARNING"):
-            warn_if_truncated(response, "self_check_input")
+            result = warn_if_truncated(response, "self_check_input")
+        assert result is True
         assert any("self_check_input" in rec.message and "length" in rec.message for rec in caplog.records)
 
     def test_silent_on_non_empty_content(self, caplog):
         from nemoguardrails.types import LLMResponse
 
         response = LLMResponse(content="yes", finish_reason="length")
         with caplog.at_level("WARNING"):
-            warn_if_truncated(response, "self_check_input")
+            result = warn_if_truncated(response, "self_check_input")
+        assert result is False
         assert not caplog.records
 
     def test_silent_on_non_length_finish_reason(self, caplog):
         from nemoguardrails.types import LLMResponse
 
         response = LLMResponse(content="", finish_reason="stop")
         with caplog.at_level("WARNING"):
-            warn_if_truncated(response, "self_check_input")
+            result = warn_if_truncated(response, "self_check_input")
+        assert result is False
+        assert not caplog.records
+
+    def test_silent_on_none_finish_reason(self, caplog):
+        from nemoguardrails.types import LLMResponse
+
+        response = LLMResponse(content="", finish_reason=None)
+        with caplog.at_level("WARNING"):
+            result = warn_if_truncated(response, "self_check_input")
+        assert result is False
         assert not caplog.records