fix(converter): preserve agent reference in pydantic coercion fallback

lib/crewai/src/crewai/agent/core.py

@@ -1436,12 +1436,13 @@ def get_code_execution_tools(self) -> list[Any]:
         )
         return []
 
-    @staticmethod
     def get_output_converter(
-        llm: BaseLLM, text: str, model: type[BaseModel], instructions: str
+        self, llm: BaseLLM, text: str, model: type[BaseModel], instructions: str
     ) -> Converter:
         """Create a Converter instance for transforming LLM output to a structured model."""
-        return Converter(llm=llm, text=text, model=model, instructions=instructions)
+        return Converter(
+            llm=llm, text=text, model=model, instructions=instructions, agent=self
+        )
 
     def _training_handler(self, task_prompt: str) -> str:
         """Handle training data for the agent task prompt to improve output on Training."""
@@ -2040,7 +2041,13 @@ def _build_output_from_result(
                         "formatted_task_instructions"
                     ).format(output_format=schema)
 
-                    converter = Converter(
+                    # Build the fallback converter through get_output_converter
+                    # so it carries the agent: the partial-JSON path can then
+                    # run the (bounded) instructions-based refit instead of
+                    # aborting with the #7984 TypeError. The refit converter
+                    # itself is still agent-less (see convert_with_instructions),
+                    # so no nested LLM call tree can form here either.
+                    converter = self.get_output_converter(
                         llm=cast(BaseLLM, self.llm),
                         text=raw_output,
                         model=response_format,

lib/crewai/src/crewai/agents/agent_builder/utilities/base_output_converter.py

@@ -32,6 +32,11 @@ class OutputConverter(BaseModel, ABC):
         description="Max number of attempts to try to get the output formatted.",
         default=3,
     )
+    agent: Any = Field(
+        default=None,
+        description="Optional reference to the calling agent, used when the "
+        "partial-JSON fallback needs to run another LLM round-trip.",
+    )
 
     @abstractmethod
     def to_pydantic(self, current_attempt: int = 1) -> BaseModel:

lib/crewai/src/crewai/utilities/converter.py

@@ -67,7 +67,7 @@ def _coerce_response_to_pydantic(self, response: Any) -> BaseModel:
                 result=response,
                 model=self.model,
                 is_json_output=False,
-                agent=None,
+                agent=self.agent,
             )
             if isinstance(partial, BaseModel):
                 return partial
@@ -405,9 +405,14 @@ def convert_with_instructions(
         raise ValueError("Agent must have a valid LLM instance for conversion")
 
     instructions = get_conversion_instructions(model=model, llm=llm)
+    # The refit converter must NOT carry the calling agent: an agent-stamped
+    # converter routes its own partial-JSON fallback back through
+    # convert_with_instructions, spawning a nested LLM call tree on every
+    # failed attempt (unbounded provider fan-out). With agent=None, a
+    # ValidationError stays inside this converter's max_attempts loop.
     converter = create_converter(
-        agent=agent,
-        converter_cls=converter_cls,
+        agent=None,
+        converter_cls=converter_cls or Converter,
         llm=llm,
         text=result,
         model=model,
@@ -549,9 +554,14 @@ async def async_convert_with_instructions(
         raise ValueError("Agent must have a valid LLM instance for conversion")
 
     instructions = get_conversion_instructions(model=model, llm=llm)
+    # The refit converter must NOT carry the calling agent: an agent-stamped
+    # converter routes its own partial-JSON fallback back through
+    # convert_with_instructions, spawning a nested LLM call tree on every
+    # failed attempt (unbounded provider fan-out). With agent=None, a
+    # ValidationError stays inside this converter's max_attempts loop.
     converter = create_converter(
-        agent=agent,
-        converter_cls=converter_cls,
+        agent=None,
+        converter_cls=converter_cls or Converter,
         llm=llm,
         text=result,
         model=model,