style: fix lint issues in test_task_output_json_overrides_llm_response_format

Applied ruff auto-fixes to remove trailing whitespace from docstring
and blank lines in the new test function.

Co-Authored-By: João <joao@crewai.com>

src/crewai/llm.py

@@ -355,12 +355,14 @@ class LLM(BaseLLM):
         self,
         messages: str | list[dict[str, str]],
         tools: list[dict] | None = None,
+        from_task: Any | None = None,
     ) -> dict[str, Any]:
         """Prepare parameters for the completion call.
 
         Args:
             messages: Input messages for the LLM
             tools: Optional list of tool schemas
+            from_task: Optional task object to check for output settings
 
         Returns:
             Dict[str, Any]: Parameters for the completion call
@@ -370,7 +372,16 @@ class LLM(BaseLLM):
             messages = [{"role": "user", "content": messages}]
         formatted_messages = self._format_messages_for_provider(messages)
 
-        # --- 2) Prepare the parameters for the completion call
+        # Task-level output settings take precedence over LLM-level response_format
+        response_format = self.response_format
+        if from_task and hasattr(from_task, 'output_json') and from_task.output_json:
+            # Task has output_json set, so ignore LLM's response_format
+            response_format = None
+        elif from_task and hasattr(from_task, 'output_pydantic') and from_task.output_pydantic:
+            # Task has output_pydantic set, so ignore LLM's response_format
+            response_format = None
+
+        # --- 3) Prepare the parameters for the completion call
         params = {
             "model": self.model,
             "messages": formatted_messages,
@@ -383,7 +394,7 @@ class LLM(BaseLLM):
             "presence_penalty": self.presence_penalty,
             "frequency_penalty": self.frequency_penalty,
             "logit_bias": self.logit_bias,
-            "response_format": self.response_format,
+            "response_format": response_format,
             "seed": self.seed,
             "logprobs": self.logprobs,
             "top_logprobs": self.top_logprobs,
@@ -1015,7 +1026,7 @@ class LLM(BaseLLM):
                 self.set_callbacks(callbacks)
             try:
                 # --- 6) Prepare parameters for the completion call
-                params = self._prepare_completion_params(messages, tools)
+                params = self._prepare_completion_params(messages, tools, from_task)
                 # --- 7) Make the completion call and handle response
                 if self.stream:
                     return self._handle_streaming_response(

tests/cassettes/test_task_output_json_overrides_llm_response_format.yaml

@@ -0,0 +1,131 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Test Agent. You''re an
+      expert in testing priority systems.\nYour personal goal is: Test goal for priority testing\nTo give my best complete final answer to the task use the exact following
+      format:\n\nThought: I now can give a great answer\nFinal Answer: Your final
+      answer must be the great and the most complete as possible, it must be outcome
+      described.\n\nI MUST use these formats, my job depends on it!"}, {"role": "user",
+      "content": "\nCurrent Task: Analyze the priority system and provide a result\n\nThis is the expect criteria
+      for your final answer: A structured result with task_result and task_confidence fields.\nyou MUST return the actual complete
+      content as the final answer, not a summary.\n\nBegin! This is VERY important
+      to you, use the tools available and give your best Final Answer, your job depends
+      on it!\n\nThought:"}], "model": "gpt-4o-mini"}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.47.0
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    content: "{\n  \"id\": \"chatcmpl-test123\",\n  \"object\":
+      \"chat.completion\",\n  \"created\": 1727214471,\n  \"model\": \"gpt-4o-mini\",\n
+      \ \"choices\": [\n    {\n      \"index\": 0,\n      \"message\": {\n        \"role\":
+      \"assistant\",\n        \"content\": \"I now can give a great answer\\nFinal
+      Answer: The priority system is working correctly with task-level settings taking precedence over agent-level settings.\",\n        \"refusal\": null\n      },\n      \"logprobs\": null,\n
+      \     \"finish_reason\": \"stop\"\n    }\n  ],\n  \"usage\": {\n    \"prompt_tokens\":
+      186,\n    \"completion_tokens\": 25,\n    \"total_tokens\": 211,\n    \"completion_tokens_details\":
+      {\n      \"reasoning_tokens\": 0\n    }\n  },\n  \"system_fingerprint\": \"fp_test\"\n}\n"
+    headers:
+      CF-Cache-Status:
+      - DYNAMIC
+      CF-RAY:
+      - 8c85f9af4ef31cf3-GRU
+      Connection:
+      - keep-alive
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 24 Sep 2024 21:47:52 GMT
+      Server:
+      - cloudflare
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '170'
+      openai-version:
+      - '2020-10-01'
+      x-ratelimit-limit-requests:
+      - '10000'
+      x-ratelimit-limit-tokens:
+      - '30000000'
+      x-ratelimit-remaining-requests:
+      - '9999'
+      x-ratelimit-remaining-tokens:
+      - '29999781'
+    http_version: HTTP/1.1
+    status_code: 200
+- request:
+    body: '{"messages": [{"role": "user", "content": "The priority system is working correctly with task-level settings taking precedence over agent-level settings."}, {"role": "system", "content":
+      "I''m gonna convert this raw text into valid JSON.\n\nThe json should have the
+      following structure, with the following keys:\n{\n    task_result: str\n    task_confidence: float\n}"}], "model":
+      "gpt-4o-mini", "tool_choice": {"type": "function", "function": {"name": "TaskOutputModel"}},
+      "tools": [{"type": "function", "function": {"name": "TaskOutputModel", "description":
+      "Correctly extracted `TaskOutputModel` with all the required parameters with correct
+      types", "parameters": {"properties": {"task_result": {"title": "Task Result", "type": "string"}, "task_confidence": {"title": "Task Confidence", "type": "number"}},
+      "required": ["task_result", "task_confidence"], "type": "object"}}}]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.47.0
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    content: "{\n  \"id\": \"chatcmpl-test456\",\n  \"object\":
+      \"chat.completion\",\n  \"created\": 1727214472,\n  \"model\": \"gpt-4o-mini\",\n
+      \ \"choices\": [\n    {\n      \"index\": 0,\n      \"message\": {\n        \"role\":
+      \"assistant\",\n        \"content\": null,\n        \"tool_calls\": [\n          {\n
+      \           \"id\": \"call_test789\",\n            \"type\":
+      \"function\",\n            \"function\": {\n              \"name\": \"TaskOutputModel\",\n
+      \             \"arguments\": \"{\\\"task_result\\\":\\\"The priority system is working correctly with task-level settings taking precedence over agent-level settings.\\\",\\\"task_confidence\\\":0.95}\"\n            }\n          }\n
+      \       ],\n        \"refusal\": null\n      },\n      \"logprobs\": null,\n
+      \     \"finish_reason\": \"stop\"\n    }\n  ],\n  \"usage\": {\n    \"prompt_tokens\":
+      120,\n    \"completion_tokens\": 30,\n    \"total_tokens\": 150,\n    \"completion_tokens_details\":
+      {\n      \"reasoning_tokens\": 0\n    }\n  },\n  \"system_fingerprint\": \"fp_test\"\n}\n"
+    headers:
+      CF-Cache-Status:
+      - DYNAMIC
+      CF-RAY:
+      - 8c85f9b2fc671cf3-GRU
+      Connection:
+      - keep-alive
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 24 Sep 2024 21:47:52 GMT
+      Server:
+      - cloudflare
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '163'
+      openai-version:
+      - '2020-10-01'
+      x-ratelimit-limit-requests:
+      - '10000'
+      x-ratelimit-limit-tokens:
+      - '30000000'
+      x-ratelimit-remaining-requests:
+      - '9999'
+      x-ratelimit-remaining-tokens:
+      - '29999947'
+    http_version: HTTP/1.1
+    status_code: 200
+version: 1

tests/test_task.py

@@ -1635,3 +1635,62 @@ def test_task_interpolation_with_hyphens():
     assert "say hello world" in task.prompt()
 
     assert result.raw == "Hello, World!"
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_task_output_json_overrides_llm_response_format():
+    """Test that task.output_json takes priority over llm.response_format when both are set.
+
+    This test addresses issue #3639: when both a task's output_json and an agent's LLM
+    response_format are set with pydantic models, the task-level setting should take
+    precedence over the agent-level setting.
+    """
+
+    from crewai.llm import LLM
+
+    class TaskOutputModel(BaseModel):
+        """Expected output model for the task."""
+        task_result: str
+        task_confidence: float
+
+    class LLMOutputModel(BaseModel):
+        """Different output model set on the LLM."""
+        llm_answer: str
+        llm_score: int
+
+    llm = LLM(model="gpt-4o-mini", response_format=LLMOutputModel)
+
+    agent = Agent(
+        role="Test Agent",
+        goal="Test goal for priority testing",
+        backstory="Test backstory for priority testing",
+        llm=llm,
+        allow_delegation=False,
+    )
+
+    task = Task(
+        description="Analyze the priority system and provide a result",
+        expected_output="A structured result with task_result and task_confidence fields",
+        agent=agent,
+        output_json=TaskOutputModel,
+    )
+
+    crew = Crew(agents=[agent], tasks=[task], process=Process.sequential)
+    result = crew.kickoff()
+
+    assert result.json_dict is not None, "Result should have json_dict output"
+    assert "task_result" in result.json_dict, (
+        "Should have task_result field from TaskOutputModel. "
+        "Task-level output_json should override LLM-level response_format."
+    )
+    assert "task_confidence" in result.json_dict, (
+        "Should have task_confidence field from TaskOutputModel"
+    )
+    assert "llm_answer" not in result.json_dict, (
+        "Should not have llm_answer field from LLMOutputModel. "
+        "This proves task-level output_json took precedence over LLM-level response_format."
+    )
+    assert "llm_score" not in result.json_dict, (
+        "Should not have llm_score field from LLMOutputModel"
+    )
+    assert isinstance(result.json_dict["task_result"], str), "task_result should be a string"
+    assert isinstance(result.json_dict["task_confidence"], (int, float)), "task_confidence should be a number"