Merge branch 'main' into lg-guardrail-llm

This parameter was used to get the model from the `task.agent` which is a quite bit redudant since we could propagate the llm directly

docs/concepts/tasks.mdx

@@ -322,6 +322,10 @@ blog_task = Task(
    - On success: it returns a tuple of `(bool, Any)`. For example: `(True, validated_result)` 
    - On Failure: it returns a tuple of `(bool, str)`. For example: `(False, "Error message explain the failure")`
 
+### TaskGuardrail
+
+The `TaskGuardrail` class offers a robust mechanism for validating task outputs
+
 ### Error Handling Best Practices
 
 1. **Structured Error Responses**:
@@ -750,6 +754,8 @@ Task guardrails provide a powerful way to validate, transform, or filter task ou
 
 ### Basic Usage
 
+#### Define your own logic to validate
+
 ```python Code
 from typing import Tuple, Union
 from crewai import Task
@@ -769,6 +775,57 @@ task = Task(
 )
 ```
 
+#### Leverage a no-code approach for validation
+
+```python Code
+from crewai import Task
+
+task = Task(
+    description="Generate JSON data",
+    expected_output="Valid JSON object",
+    guardrail="Ensure the response is a valid JSON object"
+)
+```
+
+#### Using YAML
+
+```yaml
+research_task:
+  ...
+  guardrail: make sure each bullet contains a minimum of 100 words
+  ...
+```
+
+```python Code
+@CrewBase
+class InternalCrew:
+    agents_config = "config/agents.yaml"
+    tasks_config = "config/tasks.yaml"
+
+    ...
+    @task
+    def research_task(self):
+        return Task(config=self.tasks_config["research_task"])  # type: ignore[index]
+    ...
+```
+
+
+#### Use custom models for code generation
+
+```python Code
+from crewai import Task
+from crewai.llm import LLM
+
+task = Task(
+    description="Generate JSON data",
+    expected_output="Valid JSON object",
+    guardrail=TaskGuardrail(
+        description="Ensure the response is a valid JSON object",
+        llm=LLM(model="gpt-4o-mini"),
+    )
+)
+```
+
 ### How Guardrails Work
 
 1. **Optional Attribute**: Guardrails are an optional attribute at the task level, allowing you to add validation only where needed.

src/crewai/llm.py

@@ -483,6 +483,7 @@ class LLM(BaseLLM):
                     full_response += chunk_content
 
                     # Emit the chunk event
+                    assert hasattr(crewai_event_bus, "emit")
                     crewai_event_bus.emit(
                         self,
                         event=LLMStreamChunkEvent(chunk=chunk_content),
@@ -611,6 +612,7 @@ class LLM(BaseLLM):
                 return full_response
 
             # Emit failed event and re-raise the exception
+            assert hasattr(crewai_event_bus, "emit")
             crewai_event_bus.emit(
                 self,
                 event=LLMCallFailedEvent(error=str(e)),
@@ -633,7 +635,7 @@ class LLM(BaseLLM):
                 current_tool_accumulator.function.arguments += (
                     tool_call.function.arguments
                 )
-
+            assert hasattr(crewai_event_bus, "emit")
             crewai_event_bus.emit(
                 self,
                 event=LLMStreamChunkEvent(
@@ -806,6 +808,7 @@ class LLM(BaseLLM):
                     function_name, lambda: None
                 )  # Ensure fn is always a callable
                 logging.error(f"Error executing function '{function_name}': {e}")
+                assert hasattr(crewai_event_bus, "emit")
                 crewai_event_bus.emit(
                     self,
                     event=LLMCallFailedEvent(error=f"Tool execution error: {str(e)}"),
@@ -843,6 +846,7 @@ class LLM(BaseLLM):
             LLMContextLengthExceededException: If input exceeds model's context limit
         """
         # --- 1) Emit call started event
+        assert hasattr(crewai_event_bus, "emit")
         crewai_event_bus.emit(
             self,
             event=LLMCallStartedEvent(
@@ -891,6 +895,7 @@ class LLM(BaseLLM):
                 # whether to summarize the content or abort based on the respect_context_window flag
                 raise
             except Exception as e:
+                assert hasattr(crewai_event_bus, "emit")
                 crewai_event_bus.emit(
                     self,
                     event=LLMCallFailedEvent(error=str(e)),
@@ -905,6 +910,7 @@ class LLM(BaseLLM):
             response (str): The response from the LLM call.
             call_type (str): The type of call, either "tool_call" or "llm_call".
         """
+        assert hasattr(crewai_event_bus, "emit")
         crewai_event_bus.emit(
             self,
             event=LLMCallCompletedEvent(response=response, call_type=call_type),

src/crewai/project/crew_base.py

@@ -246,6 +246,9 @@ def CrewBase(cls: T) -> T:
                     callback_functions[callback]() for callback in callbacks
                 ]
 
+            if guardrail := task_info.get("guardrail"):
+                self.tasks_config[task_name]["guardrail"] = guardrail
+
     # Include base class (qual)name in the wrapper class (qual)name.
     WrappedClass.__name__ = CrewBase.__name__ + "(" + cls.__name__ + ")"
     WrappedClass.__qualname__ = CrewBase.__qualname__ + "(" + cls.__name__ + ")"

src/crewai/task.py

@@ -140,9 +140,9 @@ class Task(BaseModel):
         default=None,
     )
     processed_by_agents: Set[str] = Field(default_factory=set)
-    guardrail: Optional[Callable[[TaskOutput], Tuple[bool, Any]]] = Field(
+    guardrail: Optional[Union[Callable[[TaskOutput], Tuple[bool, Any]], str]] = Field(
         default=None,
-        description="Function to validate task output before proceeding to next task",
+        description="Function or string description of a guardrail to validate task output before proceeding to next task",
     )
     max_retries: int = Field(
         default=3, description="Maximum number of retries when guardrail fails"
@@ -157,8 +157,12 @@ class Task(BaseModel):
 
     @field_validator("guardrail")
     @classmethod
-    def validate_guardrail_function(cls, v: Optional[Callable]) -> Optional[Callable]:
-        """Validate that the guardrail function has the correct signature and behavior.
+    def validate_guardrail_function(
+        cls, v: Optional[str | Callable]
+    ) -> Optional[str | Callable]:
+        """
+        If v is a callable, validate that the guardrail function has the correct signature and behavior.
+        If v is a string, return it as is.
 
         While type hints provide static checking, this validator ensures runtime safety by:
         1. Verifying the function accepts exactly one parameter (the TaskOutput)
@@ -171,16 +175,16 @@ class Task(BaseModel):
         - Clear error messages help users debug guardrail implementation issues
 
         Args:
-            v: The guardrail function to validate
+            v: The guardrail function to validate or a string describing the guardrail task
 
         Returns:
-            The validated guardrail function
+            The validated guardrail function or a string describing the guardrail task
 
         Raises:
             ValueError: If the function signature is invalid or return annotation
                        doesn't match Tuple[bool, Any]
         """
-        if v is not None:
+        if v is not None and callable(v):
             sig = inspect.signature(v)
             positional_args = [
                 param
@@ -211,6 +215,7 @@ class Task(BaseModel):
                     )
         return v
 
+    _guardrail: Optional[Callable] = PrivateAttr(default=None)
     _original_description: Optional[str] = PrivateAttr(default=None)
     _original_expected_output: Optional[str] = PrivateAttr(default=None)
     _original_output_file: Optional[str] = PrivateAttr(default=None)
@@ -231,6 +236,20 @@ class Task(BaseModel):
                 )
         return self
 
+    @model_validator(mode="after")
+    def ensure_guardrail_is_callable(self) -> "Task":
+        if callable(self.guardrail):
+            self._guardrail = self.guardrail
+        elif isinstance(self.guardrail, str):
+            from crewai.tasks.task_guardrail import TaskGuardrail
+
+            assert self.agent is not None
+            self._guardrail = TaskGuardrail(
+                description=self.guardrail, llm=self.agent.llm
+            )
+
+        return self
+
     @field_validator("id", mode="before")
     @classmethod
     def _deny_user_set_id(cls, v: Optional[UUID4]) -> None:
@@ -407,10 +426,8 @@ class Task(BaseModel):
                 output_format=self._get_output_format(),
             )
 
-            if self.guardrail:
-                guardrail_result = GuardrailResult.from_tuple(
-                    self.guardrail(task_output)
-                )
+            if self._guardrail:
+                guardrail_result = self._process_guardrail(task_output)
                 if not guardrail_result.success:
                     if self.retry_count >= self.max_retries:
                         raise Exception(
@@ -464,13 +481,46 @@ class Task(BaseModel):
                     )
                 )
                 self._save_file(content)
-            crewai_event_bus.emit(self, TaskCompletedEvent(output=task_output, task=self))
+            crewai_event_bus.emit(
+                self, TaskCompletedEvent(output=task_output, task=self)
+            )
             return task_output
         except Exception as e:
             self.end_time = datetime.datetime.now()
             crewai_event_bus.emit(self, TaskFailedEvent(error=str(e), task=self))
             raise e  # Re-raise the exception after emitting the event
 
+    def _process_guardrail(self, task_output: TaskOutput) -> GuardrailResult:
+        assert self._guardrail is not None
+
+        from crewai.utilities.events import (
+            TaskGuardrailCompletedEvent,
+            TaskGuardrailStartedEvent,
+        )
+        from crewai.utilities.events.crewai_event_bus import crewai_event_bus
+
+        result = self._guardrail(task_output)
+
+        crewai_event_bus.emit(
+            self,
+            TaskGuardrailStartedEvent(
+                guardrail=self._guardrail, retry_count=self.retry_count
+            ),
+        )
+
+        guardrail_result = GuardrailResult.from_tuple(result)
+
+        crewai_event_bus.emit(
+            self,
+            TaskGuardrailCompletedEvent(
+                success=guardrail_result.success,
+                result=guardrail_result.result,
+                error=guardrail_result.error,
+                retry_count=self.retry_count,
+            ),
+        )
+        return guardrail_result
+
     def prompt(self) -> str:
         """Prompt the task.
 

src/crewai/tasks/task_guardrail.py

@@ -0,0 +1,92 @@
+from typing import Any, Optional, Tuple
+
+from pydantic import BaseModel, Field
+
+from crewai.agent import Agent, LiteAgentOutput
+from crewai.llm import LLM
+from crewai.task import Task
+from crewai.tasks.task_output import TaskOutput
+
+
+class TaskGuardrailResult(BaseModel):
+    valid: bool = Field(
+        description="Whether the task output complies with the guardrail"
+    )
+    feedback: str | None = Field(
+        description="A feedback about the task output if it is not valid",
+        default=None,
+    )
+
+
+class TaskGuardrail:
+    """It validates the output of another task using an LLM.
+
+    This class is used to validate the output from a Task based on specified criteria.
+    It uses an LLM to validate the output and provides a feedback if the output is not valid.
+
+    Args:
+        description (str): The description of the validation criteria.
+        llm (LLM, optional): The language model to use for code generation.
+    """
+
+    def __init__(
+        self,
+        description: str,
+        llm: LLM,
+    ):
+        self.description = description
+
+        self.llm: LLM = llm
+
+    def _validate_output(self, task_output: TaskOutput) -> LiteAgentOutput:
+        agent = Agent(
+            role="Guardrail Agent",
+            goal="Validate the output of the task",
+            backstory="You are a expert at validating the output of a task. By providing effective feedback if the output is not valid.",
+            llm=self.llm,
+        )
+
+        query = f"""
+        Ensure the following task result complies with the given guardrail.
+
+        Task result:
+        {task_output.raw}
+
+        Guardrail:
+        {self.description}
+        
+        Your task:
+        - Confirm if the Task result complies with the guardrail.
+        - If not, provide clear feedback explaining what is wrong (e.g., by how much it violates the rule, or what specific part fails).
+        - Focus only on identifying issues — do not propose corrections.
+        - If the Task result complies with the guardrail, saying that is valid
+        """
+
+        result = agent.kickoff(query, response_format=TaskGuardrailResult)
+
+        return result
+
+    def __call__(self, task_output: TaskOutput) -> Tuple[bool, Any]:
+        """Validates the output of a task based on specified criteria.
+
+        Args:
+            task_output (TaskOutput): The output to be validated.
+
+        Returns:
+            Tuple[bool, Any]: A tuple containing:
+                - bool: True if validation passed, False otherwise
+                - Any: The validation result or error message
+        """
+
+        try:
+            result = self._validate_output(task_output)
+            assert isinstance(
+                result.pydantic, TaskGuardrailResult
+            ), "The guardrail result is not a valid pydantic model"
+
+            if result.pydantic.valid:
+                return True, task_output.raw
+            else:
+                return False, result.pydantic.feedback
+        except Exception as e:
+            return False, f"Error while validating the task output: {str(e)}"

src/crewai/utilities/events/__init__.py

@@ -9,6 +9,10 @@ from .crew_events import (
     CrewTestCompletedEvent,
     CrewTestFailedEvent,
 )
+from .task_guardrail_events import (
+    TaskGuardrailCompletedEvent,
+    TaskGuardrailStartedEvent,
+)
 from .agent_events import (
     AgentExecutionStartedEvent,
     AgentExecutionCompletedEvent,

src/crewai/utilities/events/event_types.py

@@ -34,6 +34,10 @@ from .task_events import (
     TaskFailedEvent,
     TaskStartedEvent,
 )
+from .task_guardrail_events import (
+    TaskGuardrailCompletedEvent,
+    TaskGuardrailStartedEvent,
+)
 from .tool_usage_events import (
     ToolUsageErrorEvent,
     ToolUsageFinishedEvent,
@@ -68,4 +72,6 @@ EventTypes = Union[
     LLMCallCompletedEvent,
     LLMCallFailedEvent,
     LLMStreamChunkEvent,
+    TaskGuardrailStartedEvent,
+    TaskGuardrailCompletedEvent,
 ]

src/crewai/utilities/events/task_guardrail_events.py

@@ -0,0 +1,38 @@
+from typing import Any, Callable, Optional, Union
+
+from crewai.utilities.events.base_events import BaseEvent
+
+
+class TaskGuardrailStartedEvent(BaseEvent):
+    """Event emitted when a guardrail task starts
+
+    Attributes:
+        guardrail: The guardrail callable or TaskGuardrail instance
+        retry_count: The number of times the guardrail has been retried
+    """
+
+    type: str = "task_guardrail_started"
+    guardrail: Union[str, Callable]
+    retry_count: int
+
+    def __init__(self, **data):
+        from inspect import getsource
+
+        from crewai.tasks.task_guardrail import TaskGuardrail
+
+        super().__init__(**data)
+
+        if isinstance(self.guardrail, TaskGuardrail):
+            self.guardrail = self.guardrail.description.strip()
+        elif isinstance(self.guardrail, Callable):
+            self.guardrail = getsource(self.guardrail).strip()
+
+
+class TaskGuardrailCompletedEvent(BaseEvent):
+    """Event emitted when a guardrail task completes"""
+
+    type: str = "task_guardrail_completed"
+    success: bool
+    result: Any
+    error: Optional[str] = None
+    retry_count: int

tests/cassettes/test_guardrail_emits_events.yaml

@@ -0,0 +1,713 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Test Agent. Test Backstory\nYour
+      personal goal is: Test Goal\nTo give my best complete final answer to the task
+      respond using 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: Gather information
+      about available books on the First World War\n\nThis is the expected criteria
+      for your final answer: A list of available books on the First World War\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", "stop":
+      ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '903'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFbbbhs3EH33Vwz2pUAgGXbqxInenJtzgZs0dmKgdWHMcmd3J+ZytkOu
+        ZLUI0N/o7/VLCpK7kpwL2hfL0pDDM2fOXP7cAyi4KhZQmBaD6Xo7f/Ke3rz75G+ffXz04celD+ef
+        yifL8hc5P/145opZvCHlJzJhurVvpOstBZbRbJQwUPR6eHz04PH9xw+Oj5Ohk4psvNb0YX4k844d
+        z+8f3D+aHxzPDx+Nt1thQ75YwK97AAB/pr8Rp6votljAwWz6pSPvsaFisTkEUKjY+EuB3rMP6EIx
+        2xqNuEAuQX8FTlZg0EHDSwKEJsIGdH5FCnDlXrBDCyfp+wJekhKwBwTLPoDU4CRgaQlKkRsP4iC0
+        BC9YfYBLUVvBJeoMyEV6IhjXwBKVZfDQk/qeTOAleUBXxasd+cWVu3KH+3Dv3lVxYi38PDCFyfMl
+        +UDq4IWKC1cFlGt4rmxaOENlhPfUof4+0L17MIcTcLIkCxX1bEJ8OXqg256UyRnyEf8paYcOvNiK
+        ST1Ug8aTG+zwagYtN63lpt34aEVVNN0PSs60sEKtUWk/Yr+fsV98TUQG/FpaB2+IGnQjzkiOUkvO
+        xyTIknTJtEruW4q+Z4AO7fqPCICDB4ODJz+DDj+JQokh2Pg1kmjEefp9SAEmOD+OVI4wPrhKHC0g
+        wjsPouvpmdOU+ZSuw8eHRxAkfj7KmE/3X+/DGa1JJ2pRFWPqoGWf3IQWA5gI3meiXWCliKe2bAKs
+        OLSAUIsZklBuaA20JBdy8mtuBh0hH20ZPB1c4vlkaAY/ZvwJaomKcLkPF4Npu4lHB2iMRe6ogpXo
+        DVQUkO0m8/kxS1jFn4Y+hjgSPOkPCNWuwQdssj6MdCWGGVDXt+j5j8lZRYY9SwYXXZJm7A8y9lOR
+        qlxTfCKK+KLFEfx7KUkDnCouyW9wD0FKlkaxbycmsQ+RkC9E+7XmoFbpACcJ/3CnsGaAVlyTyY+e
+        gqLzHNtUhNaLD/MYvuU6i/dhhn8Zz78dFJ6gufETUZdo7QI+skkJj5Q9ozqqht2OVp7hkis4D5Fv
+        L9vcRAV79pPe6tRZ1oR6R+gyBMsu0lzKiLljywE1pkUxUMPbfgFeDMu8F8uBDVpIve02+Eyib7Gn
+        fFCGYKTLQR7v6GtS/QJO4GlKNrwcFT3C+mYRv6NACi9RQ4rvomWfeiD0KkuuIsKxLuflep7/i/KU
+        wYUZDKkRxjxJJKFi1Cmo8UzivNRJbiuMiR0VHGSbr0fbUGK3CgLPXZUkd4nqc5mfdKQce/zZROTz
+        rZ7Y7Ta7saVWK9QKzvbhqdT1Tn3RbW8l1r24VJSTY3ZLsUvqyCUtbJJZsTdD7vrfSyJ3PZqQa+dx
+        DubtkjT5uJA+puU8K/ufv/728IzxPzJzoqEdFJ53Pa3HZlXHc+0Ou9+oopVyCOSiB3TbyJZiBxeI
+        ojLrmlJCEBRX01jJTOQ2sTtcUkCHB9+dBTGwn2gFr6wdMiXVJLyx5YpW4uCpqPJ2VMRhZ9fAO7eS
+        7JLceyWfGyqYVsWJlSZVRXtX0XmiGEPecxzetSho7mHxCFoLJZqbRmVwVY4kCTwN/R2Bx0lOIbll
+        5+OEzDndDAd/p5elt/OedMshNxDPjeM6Uj32tq+Z2t/dXZTqwWPcn9xg7Y4BXVxFYj7S1vTbaPm8
+        2ZOsNL1K6b+4WtTs2LfXSujFxZ3IB+mLZP28B/Bb2seGOytW0at0fbgOckPpucPjcR8rtmvg1nqU
+        Nz+AIkhAuzU8PJoMdxxe57nld1a6wqBpqdpe3e5/OFQsO4a9nbC/hvMt3zl0ds3/cb81GEN9oOq6
+        V6pyJXzrmFJck793bENzAlz4uPkYug5MGlNRUY2Dzctr4dc+UHdds2tIe+W8wdb9dVWiwYcHVX1Q
+        7H3e+xcAAP//AwDV5F0jzwsAAA==
+    headers:
+      CF-RAY:
+      - 937ec970ab837df5-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:26:28 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=lSQwOucDEe7UVr7Rj6WZvsTkBkgGT9q7hgVX9AzK42s-1745929588-1.0.1.1-6m0TeDAKI0Hgbl6.GmWHMBMkIpmnfhOu3jQKfjmcvWLHqWUWoE1O4xa9VCtZYXv6_9poUVQq_oCNtzy8eL1XDc8_J3aRMOG3LCvOyvqCawk;
+        path=/; expires=Tue, 29-Apr-25 12:56:28 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=TeUS2y0sO.6zigvKZd5.0zRS7sujoYCd9.wMQJboxxo-1745929588265-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '10782'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999808'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_cbf6ad64d8b470c6d9eff91852f85e1c
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Guardrail Agent. You
+      are a expert at validating the output of a task. By providing effective feedback
+      if the output is not valid.\nYour personal goal is: Validate the output of the
+      task\n\nTo give my best complete final answer to the task respond using 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!\nIMPORTANT:
+      Your final answer MUST contain all the information requested in the following
+      format: {\n  \"valid\": bool,\n  \"feedback\": str | None\n}\n\nIMPORTANT: Ensure
+      the final output does not include any code block markers like ```json or ```python."},
+      {"role": "user", "content": "\n        Ensure the following task result complies
+      with the given guardrail.\n\n        Task result:\n        Here is a list of
+      notable books on the First World War, encompassing various perspectives and
+      themes:\n\n1. **\"All Quiet on the Western Front\" by Erich Maria Remarque**
+      - A novel depicting the experiences of German soldiers during World War I, highlighting
+      the horrors of trench warfare.\n\n2. **\"The First World War\" by John Keegan**
+      - A comprehensive overview of the war, analyzing its causes, major battles,
+      and consequences.\n\n3. **\"A World Undone: The Story of the Great War, 1914
+      to 1918\" by G.J. Meyer** - A narrative history that covers the entire conflict
+      with a focus on key events and figures.\n\n4. **\"The Guns of August\" by Barbara
+      W. Tuchman** - An acclaimed work detailing the events leading up to the war
+      and the early stages of combat, emphasizing the decisions of leaders.\n\n5.
+      **\"Goodbye to All That\" by Robert Graves** - An autobiography that captures
+      the experience of trench warfare from a soldier''s perspective, along with the
+      transition to post-war life.\n\n6. **\"With Our Backs to the Wall: Victory and
+      Defeat in 1918\" by David Stevenson** - An analysis of the final year of the
+      war, outlining both the military strategies and the socio-political contexts
+      that shaped the outcome.\n\n7. **\"The Great War: A Combat History of the First
+      World War\" by Peter Hart** - This book provides a battle-by-battle account,
+      using personal diaries and accounts to bring the war''s events to life.\n\n8.
+      **\"The War to End All Wars: The American Military Experience in World War I\"
+      by Edward M. Coffman** - An exploration of American involvement in the war,
+      discussing military strategies and impacts.\n\n9. **\"Over the Top: A Soldier\u2019s
+      Diary of the First World War\" by Arthur Empey** - A firsthand account of trench
+      warfare written by an American volunteer, offering a raw depiction of combat
+      experiences.\n\n10. **\"The First World War: A New Illustrated History\" by
+      Gordon Corrigan** - A richly illustrated book that presents a chronological
+      history of the war, accessible for readers of all backgrounds.\n\nThis list
+      provides a variety of insights and narratives that capture the complexity and
+      significance of the First World War.\n\n        Guardrail:\n        Ensure the
+      authors are from Italy\n        \n        Your task:\n        - Confirm if the
+      Task result complies with the guardrail.\n        - If not, provide clear feedback
+      explaining what is wrong (e.g., by how much it violates the rule, or what specific
+      part fails).\n        - Focus only on identifying issues \u2014 do not propose
+      corrections.\n        - If the Task result complies with the guardrail, saying
+      that is valid\n        "}], "model": "gpt-4o-mini", "stop": ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '3636'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFPJbhsxDL37Kwidx4HjOrDjW4qiRZBb4m6oA4OWODNqNNREouwYgf+9
+        0HhNF6CXEYaPyyP5+NoDUNaoKShdo+imdf3393SH+qv9/vLlefXw0MZw/222nn1No9sPQRU5wi9/
+        kpZD1IX2TetIrOcdrAOhUM56OR5dXQ+vryaTDmi8IZfDqlb6I99vLNv+cDAc9Qfj/uVkH117qymq
+        KfzoAQC8dt/Mkw29qCkMioOloRixIjU9OgGo4F22KIzRRkEWVZxA7VmIO+qz2qeqlincAvs1aGSo
+        7IoAocr8ATmuKQDM+aNldHDT/U/hdc4Ac7VCZ81cTaFEF6nYGUsis0T9lO1zNasJBOMTBIrJCRhP
+        EdgLdAPbwNpKDVITVAmDCWgdLEljigTsmcCXHepsFDKASWofImAgKINv4FbQbS7gxrkj1hDnNWTv
+        g5exZUmBOFdNLMFSLMCydslYruAThQZ5U3SVPt/t34ebApANeKkpxALWtdU1rKx3KBQ7n0DPyQbK
+        FUHqPK8DhxQFltTRs8gXczXn7fkOApUpYtYBJ+fOAGT2grmBbvuPe2R73LfzVRv8Mv4WqkrLNtaL
+        QBg9591G8a3q0G0P4LHTVXojFdUG37SyEP9EXbnxaK8rdZLzGTreg+IF3ck+uTwAb/ItDAlaF8+U
+        qTTqmswp9CRjTMb6M6B31vWfbP6We9e55ep/0p8ArakVMos2kLH6bccnt0D52v/ldpxyR1hFCiur
+        aSGWQt6EoRKT292gipso1CxKyxWFNtjdIZbtYvDuejgZDgfXA9Xb9n4BAAD//wMAG/c/OJYEAAA=
+    headers:
+      CF-RAY:
+      - 937ec9b72e717e0f-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:26:30 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=2Kv7JW0NnxXgvGH0KQnR1VcNF47nEYxhEB4bYlFvRfA-1745929590-1.0.1.1-gmQvNlZR_WoNPIkf.07YchEy.a13FU2sP7wZropzue.84PKf6DVY_clzo6DqmBOk8PtKtM3gV952vhWHuE0ygk7096hW9jQTST91FDi30Jc;
+        path=/; expires=Tue, 29-Apr-25 12:56:30 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=HHxY6s22oZkfAHAh8HGGIeDkKOuavigZ5DkKL6.WyGc-1745929590485-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '1744'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999140'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_74247d0bd52dcebe834bd1cdb4b38470
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Test Agent. Test Backstory\nYour
+      personal goal is: Test Goal\nTo give my best complete final answer to the task
+      respond using 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: Gather information
+      about available books on the First World War\n\nThis is the expected criteria
+      for your final answer: A list of available books on the First World War\nyou
+      MUST return the actual complete content as the final answer, not a summary.\n\nThis
+      is the context you''re working with:\n### Previous attempt failed validation:
+      The task result does not comply with the guardrail because none of the listed
+      authors are from Italy. All authors mentioned are from different countries,
+      including Germany, the UK, the USA, and others, which violates the requirement
+      that authors must be Italian.\n\n\n### Previous result:\nHere is a list of notable
+      books on the First World War, encompassing various perspectives and themes:\n\n1.
+      **\"All Quiet on the Western Front\" by Erich Maria Remarque** - A novel depicting
+      the experiences of German soldiers during World War I, highlighting the horrors
+      of trench warfare.\n\n2. **\"The First World War\" by John Keegan** - A comprehensive
+      overview of the war, analyzing its causes, major battles, and consequences.\n\n3.
+      **\"A World Undone: The Story of the Great War, 1914 to 1918\" by G.J. Meyer**
+      - A narrative history that covers the entire conflict with a focus on key events
+      and figures.\n\n4. **\"The Guns of August\" by Barbara W. Tuchman** - An acclaimed
+      work detailing the events leading up to the war and the early stages of combat,
+      emphasizing the decisions of leaders.\n\n5. **\"Goodbye to All That\" by Robert
+      Graves** - An autobiography that captures the experience of trench warfare from
+      a soldier''s perspective, along with the transition to post-war life.\n\n6.
+      **\"With Our Backs to the Wall: Victory and Defeat in 1918\" by David Stevenson**
+      - An analysis of the final year of the war, outlining both the military strategies
+      and the socio-political contexts that shaped the outcome.\n\n7. **\"The Great
+      War: A Combat History of the First World War\" by Peter Hart** - This book provides
+      a battle-by-battle account, using personal diaries and accounts to bring the
+      war''s events to life.\n\n8. **\"The War to End All Wars: The American Military
+      Experience in World War I\" by Edward M. Coffman** - An exploration of American
+      involvement in the war, discussing military strategies and impacts.\n\n9. **\"Over
+      the Top: A Soldier\u2019s Diary of the First World War\" by Arthur Empey** -
+      A firsthand account of trench warfare written by an American volunteer, offering
+      a raw depiction of combat experiences.\n\n10. **\"The First World War: A New
+      Illustrated History\" by Gordon Corrigan** - A richly illustrated book that
+      presents a chronological history of the war, accessible for readers of all backgrounds.\n\nThis
+      list provides a variety of insights and narratives that capture the complexity
+      and significance of the First World War.\n\n\nTry again, making sure to address
+      the validation error.\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", "stop": ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '3509'
+      content-type:
+      - application/json
+      cookie:
+      - __cf_bm=2Kv7JW0NnxXgvGH0KQnR1VcNF47nEYxhEB4bYlFvRfA-1745929590-1.0.1.1-gmQvNlZR_WoNPIkf.07YchEy.a13FU2sP7wZropzue.84PKf6DVY_clzo6DqmBOk8PtKtM3gV952vhWHuE0ygk7096hW9jQTST91FDi30Jc;
+        _cfuvid=HHxY6s22oZkfAHAh8HGGIeDkKOuavigZ5DkKL6.WyGc-1745929590485-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFZtbtxGEv2vUxT4ZwFjRtAocmTpn7yBHCW24cTGGkFkCDXdRbLsZhVT
+        3ZzxKDCw19gb5B65yZ5k0U3OhxUF2D8zIJtdXa/eq9f1+xFAxb66hMq1mFzXh/nzn+lHt/rl5Q+v
+        vVu8ufrB/xTf6Q3+/Ozdx9BXs7xDlx/Jpe2uY6ddHyixyrjsjDBRjro4P3t6cXrx9OKkLHTqKeRt
+        TZ/mZzrvWHh+enJ6Nj85ny+eTbtbZUexuoRfjwAAfi+/OU/x9Lm6hBKrvOkoRmyoutx9BFCZhvym
+        whg5JpRUzfaLTiWRlNRvQHQNDgUaXhEgNDltQIlrMoBbuWbBAFfl+RK+JyPgCAiBYwKtQTThMhAs
+        VT9FUIHUElyzxQTv1YKH92iwNk6JBJYbuEkYGAVwSK1anAGLU+vVMLE0sEJjHSL0ZLEnl3hFEVB8
+        DttRvLyVW1kcw5Mnt9VNgI4VRDuCP/8AoRgH0dsqn/JdDgPX+uQJzOEKBuHfBgJBy8esCFgSWVqz
+        5DPzWZpRxqTG03ktlyeHAWp0KQIudUgFHXd9eaP1AcSbjH0LLqpjSpvjnO3pmO1LhGYgMwRPDKIx
+        Wf4T4THjVyVjXGGTI2jJ4RWaU3iDoWMyLljetRxLraE3XbHP2YKnhBzIAzqngxRecqL0uSdjEkcl
+        2X12wTNZBD9Yxv8IYzOo1Q0xr46UskGniVeYBT5WKCYbmiZQLDC/GWG+HUIgSIbOEXgKAeGFoXiC
+        FwX9CPYtWcMKbxOybDlqDPuWHYiuKEBqMcEykPgIaIljYpfxGMXIKrDm1B5ytKc2KXjq2Y1cBa4p
+        Y99h5lGhyUhcO9YlP78osn+PVsCcTZz9o+gER2HghOeNcYcTHHil4hkDjbhe8BCp7wm+I3g5JOY4
+        gcvmYNSSxJyirshWTOsRJX3ugxrFkkdD2mvgVEA5HCLFGXT4UQ2WmFLIj7n4QaWZJ7IOqK5pUmMO
+        sEbbKnFTvlzSRsUXWE93UvyKk1yTa+MhTFJ8gxoU/okpoQ339/hAeNT1LUa+nzI2arh0z9gWD7ti
+        F3sGLTdt4KZNW9FFboRrdihpi27b6w+1G7QUhFecJTwJcOK0YPt2wkYpZY/yDMOodEyTJ1yFbJTi
+        TeH6GJ4bNyiJdgSFkO1GpZxWosRD38LtcYdds86dMjZifoV7K8n5sSTuMNGhmW0d0ggzyyM4p90S
+        U4FxPsK4ur8fzLio1Vgc4aFL3EQhadqtI9DEy2fsWCZW9uV7tPOnLqhNJUHIux7reDeENBjNoFPD
+        QLMdObFFIw8OOzT0+SDATnNRTLUfCXn2F99jeIUp0UTHtaG4LDPzLDzRcOi6PNJReqTWEHSdvW6j
+        gzQ7QB3Kf//9nwgfdTChDaTWdGjaPTla11ToYolZehl60rFE2dB2ZPVx41oN2pTDDyzeqdSB3UjP
+        xQhqcbF4Ol9cLJ4dw8tHDe5KvBHC62xJW2wxx0cLG1irfRpxoWDYbDspJsNEzfYGSuRa2WWEfoXi
+        qCNJMbdg0A35wwu1Vsu98veeXgAsTnaXZ8NqopOprTgVjxvT/1eWnNznu+eQHYf9eAVIM7kWbjum
+        qKmMCrnixeYiBMLSF9uCiz9wqdkk2PxBq9kLaZOF1ZMf3asQKzhxxJ4k8XSnFjMqE8jOU7I2DKX5
+        Su/TkAHrVkclgOcVWTwYBcZa96a1DqVnH8qkG0LiGh0l8l+5UrFhatDx/hp5WPLDqcuoHiLmyU+G
+        EA4WUPIQVS7WPO99mFa+7Ca8oE1vuowPtlY1C8f2zgijSp7mYtK+KqtfjgA+lEly+Go4rHrTrk93
+        ST9ROe58cTbGq/YD7H717OJ8Wk2aMOwXFqeLxeyRiHfjLBIPptHKoWvJ7/fuR1ccPOvBwtEB7r/m
+        81jsETtL8/+E3y84R30if9cb+dFnHvvMKE/4f/fZrs4l4SrmO93RXWKyzIWnGocwzt1V3MRE3V3N
+        0pD1xuPwXfd3fokOvz3x9Ul19OXofwAAAP//AwDXPLUkigwAAA==
+    headers:
+      CF-RAY:
+      - 937ec9c4b8307e0f-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:26:41 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '10386'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999169'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_7e9c58909cf39af106866bae0c0cf7e1
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Guardrail Agent. You
+      are a expert at validating the output of a task. By providing effective feedback
+      if the output is not valid.\nYour personal goal is: Validate the output of the
+      task\n\nTo give my best complete final answer to the task respond using 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!\nIMPORTANT:
+      Your final answer MUST contain all the information requested in the following
+      format: {\n  \"valid\": bool,\n  \"feedback\": str | None\n}\n\nIMPORTANT: Ensure
+      the final output does not include any code block markers like ```json or ```python."},
+      {"role": "user", "content": "\n        Ensure the following task result complies
+      with the given guardrail.\n\n        Task result:\n        Here is a list of
+      notable books on the First World War written by Italian authors, incorporating
+      various perspectives and themes:\n\n1. **\"Il mio nome \u00e8 nessuno\" by Dario
+      Fo** - A unique narrative intertwining personal stories and historical facts
+      about the impacts of World War I on Italian society.\n\n2. **\"La guerra dei
+      nostri nonni\" by Mario avagliano and Marco Palmieri** - This book provides
+      a detailed account of the experiences of Italian soldiers during the First World
+      War, focusing on their motivations and struggles.\n\n3. **\"Sulle tracce della
+      Grande Guerra\" by Sergio Staino** - A graphic novel that blends artistic expression
+      with historical narrative to depict the life of soldiers in the trenches of
+      the Great War.\n\n4. **\"L''intera storia della Prima Guerra Mondiale\" by Giuseppe
+      De Lutiis** - A comprehensive overview that explores the geopolitical causes,
+      major battles, and long-term effects of the war on Italy and beyond.\n\n5. **\"La
+      Grande Guerra in Friuli\" by Paolo Cattaruzza** - This book emphasizes the regional
+      impact of World War I in Friuli, highlighting the significant battles and the
+      experiences of local civilians and soldiers.\n\n6. **\"Lettere di un soldato\"
+      by Alessandro F. Brigante** - A collection of letters written by a soldier during
+      the war, providing a personal and intimate perspective on the realities of combat.\n\n7.
+      **\"Azzurri in trincea\" by Mario Isnenghi** - The book examines the experience
+      of Italian soldiers in the front lines, focusing on their culture, morale, and
+      the shared camaraderie among troops.\n\n8. **\"La guerra di Matteo\" by Franco
+      Cardini** - A historical fiction that follows a young Italian man\u2019s journey
+      through the war, offering insights into the emotional and psychological impacts
+      of conflict.\n\n9. **\"1915-1918. La Grande Guerra\" by Andrea Nativi** - A
+      scholarly work that analyzes the strategies and technological advancements employed
+      by Italian forces during the First World War.\n\n10. **\"Il giorno della vittoria\"
+      by Vincenzo Pardini** - A captivating exploration of the final offensives leading
+      to the end of the war, examining how they shaped Italy\u2019s national identity.\n\nThis
+      list highlights a range of Italian authors who offer diverse narratives and
+      profound insights into the multifaceted experiences and legacies of the First
+      World War.\n\n        Guardrail:\n        Ensure the authors are from Italy\n        \n        Your
+      task:\n        - Confirm if the Task result complies with the guardrail.\n        -
+      If not, provide clear feedback explaining what is wrong (e.g., by how much it
+      violates the rule, or what specific part fails).\n        - Focus only on identifying
+      issues \u2014 do not propose corrections.\n        - If the Task result complies
+      with the guardrail, saying that is valid\n        "}], "model": "gpt-4o-mini",
+      "stop": ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '3829'
+      content-type:
+      - application/json
+      cookie:
+      - __cf_bm=2Kv7JW0NnxXgvGH0KQnR1VcNF47nEYxhEB4bYlFvRfA-1745929590-1.0.1.1-gmQvNlZR_WoNPIkf.07YchEy.a13FU2sP7wZropzue.84PKf6DVY_clzo6DqmBOk8PtKtM3gV952vhWHuE0ygk7096hW9jQTST91FDi30Jc;
+        _cfuvid=HHxY6s22oZkfAHAh8HGGIeDkKOuavigZ5DkKL6.WyGc-1745929590485-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jJJNb9swDIbv/hWEzvHgeFmT+NYNKFbstGLADnNhKBJtq5EpQZLTFUH+
+        +yDHiZ19ALsYMB++FF+SxwSAKckKYKLlQXRWpx+f8Av5dXj8vH16kG+f9j6Xuxf5vf56nx3YIirM
+        7gVFuKjeCdNZjUEZOmPhkAeMVZfr1Ydtvr3LlgPojEQdZY0N6cqknSKV5lm+SrN1utyM6tYogZ4V
+        8CMBADgO39gnSfzJCsgWl0iH3vMGWXFNAmDO6Bhh3HvlA6fAFhMUhgLS0Pq31vRNGwp4BDKvIDhB
+        ow4IHJrYP3Dyr+gASnpQxDXcD/8FHEsCKNmBayVLVkBwPS7OsRpR7rjYxzD1Wpd0mj/usO491yOc
+        AU5kAo8DHGw/j+R0NapNY53Z+d+krFakfFs55N5QNOWDsWygpwTgeRhofzMjZp3pbKiC2ePw3Ppu
+        HCib9jjRfDPCYALXM9XmAm7qVRIDV9rPVsIEFy3KSTrtj/dSmRlIZq7/7OZvtc/OFTX/U34CQqAN
+        KCvrUCpx63hKcxjP/F9p1ykPDTOP7qAEVkGhi5uQWPNen4+P+TcfsKtqRQ0669T5AmtbZe+3+SbP
+        s23GklPyCwAA//8DANS97/6PAwAA
+    headers:
+      CF-RAY:
+      - 937eca0769737e0f-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:26:42 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '924'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999094'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_242f5797b2a0bf5867b01ab6027921fa
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Test Agent. Test Backstory\nYour
+      personal goal is: Test Goal\nTo give my best complete final answer to the task
+      respond using 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: Test task\n\nThis
+      is the expected criteria for your final answer: Output\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", "stop": ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '807'
+      content-type:
+      - application/json
+      cookie:
+      - __cf_bm=2Kv7JW0NnxXgvGH0KQnR1VcNF47nEYxhEB4bYlFvRfA-1745929590-1.0.1.1-gmQvNlZR_WoNPIkf.07YchEy.a13FU2sP7wZropzue.84PKf6DVY_clzo6DqmBOk8PtKtM3gV952vhWHuE0ygk7096hW9jQTST91FDi30Jc;
+        _cfuvid=HHxY6s22oZkfAHAh8HGGIeDkKOuavigZ5DkKL6.WyGc-1745929590485-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFNNTxsxEL3nV4x8TqIQICm5tYeqiAMSqoSqFq0ce3Z3wOvZ2rMbAuK/
+        V/YGNrQcell5583He+Pn5wmAIqs2oEytxTStm325wSuOVw+if3wz14Iu7tZP29vLm+vb05Wapgre
+        3qOR16q54aZ1KMR+gE1ALZi6nqzPzi+WF6vFMgMNW3SprGpldsazhjzNlovl2Wyxnp18OlTXTAaj
+        2sDPCQDAc/4mnt7io9rAYvoaaTBGXaHavCUBqMAuRZSOkaJoL2o6goa9oM/UL8HzDoz2UFGPoKFK
+        tEH7uMMA8Mt/Ja8dfM7/G/heI5guBPQCouMDkO/Z9RihDdyTJV+BBqk5cFfVoL0Fi6LJoYWAsWUf
+        EbY6ogX2IDUCPrZoBC2YQIKBNDTo0xbRzuESduQctAFjnpimD2tOhywCdiQ1dwKxaxod6Il8NQX0
+        sQuJjNRaAHsMe9AxTQKKYLjHgBaEoUEc+gb83VHANDwCliUaoR7dfp41cyeGG3ylk6QiGKcDyT6r
+        TLQC1uhjWiP5kkOjk4yBQdm5kpyLedRgnJTHZQ6kTc6PLyhg2UWdTOI7544A7T1L7putcXdAXt7M
+        4LhqA2/jX6WqJE+xLgLqyD5dfBRuVUZfJgB32XTdOx+pNnDTSiH8gHncyfnp0E+NXh/R1foACot2
+        Y3y5PFj1fb9i8EU8sq0y2tRox9LR47qzxEfA5Ej1v2w+6j0oJ1/9T/sRMAZbQVu0AS2Z94rHtID3
+        2cQfp71tORNWEUNPBgshDOkmLJa6c8MDVXEfBZuiJF9haAMNr7RsC7vVRq8WtlyoycvkDwAAAP//
+        AwC4yvtmswQAAA==
+    headers:
+      CF-RAY:
+      - 937eca0f18177e0f-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:26:44 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '1988'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999832'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_99b127ab59c932c0b46dbecc738df8ba
+    status:
+      code: 200
+      message: OK
+version: 1

tests/cassettes/test_guardrail_when_an_error_occurs.yaml

@@ -0,0 +1,288 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Test Agent. Test Backstory\nYour
+      personal goal is: Test Goal\nTo give my best complete final answer to the task
+      respond using 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: Gather information
+      about available books on the First World War\n\nThis is the expected criteria
+      for your final answer: A list of available books on the First World War\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", "stop":
+      ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '903'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFdhb9w2Ev3uXzHQlwCBbewmThzvN1+Q2G3PQM+X1mjPhTFLjqSpKY46
+        pHazV+S/H0hK2t00Be6LYZEi9d68x8fZP08AKrbVCirTYjRd787+cU/3P7+9tfTPq+1PTffDtr7+
+        tf759+HG9tdSnaYVsv6dTJxWnRvpekeRxZdpo4SR0q7Ly4s3V68Xi+VlnujEkkvLmj6eXchZx57P
+        Xi1eXZwtLs+W78bVrbChUK3gPycAAH/mvwmnt/S5WsHidBrpKARsqFrNLwFUKi6NVBgCh4g+Vqf7
+        SSM+ks/QvwMvWzDooeENAUKTYAP6sCUFePQf2aOD6/y8gltSAg6AkOgqteRDWuc4RJAacIPscO0I
+        1iLPAcRDbAk+soYID6LOwgPq6tE/+uU5vHz5WH1qCf7tiPotumfSsIJb2cKHQaUneCAfIUpaA+xh
+        ebW8eKxgvYP3rXKI0rek8N6hPr98mcACwKeWQ/44WHIbCsA+SgZRBPoMNZooGiC2GMGRhXFehrhW
+        wufEIz1vUU9B6pqUfQPsAzdtPNivF8eRDTpAbyGIYXRgdx47NiFtQqhuB68WsQVDPg66G3mdJ/6v
+        Cv/rsSw/eSueVpDrEUV3E4ybLMhDApP4J7TLq+W7Uoeb8+/P4Y52pHMB8tOLAPS5xyKOR1WM6T8j
+        G8rMCchH1kwSthxbQKjFDFmxtcQWOnYcUXcQomKkhilknmltO3To0wdImbyhAOTtoGQTpCDOcvpK
+        etvwhh2jD4COnwvx1yNx5+BfA1OcTPJAIZJ6+KjiY6H3Qdm0cIfKCPfUof4x0Ez0GnrhxqOP4GVD
+        rghqsI+DUiGpFNhliAW74tDhAfAM+IY00Zlx2yELPrsVvjuFNQayE1AcYiupxLI9qkJmd7G39Veu
+        L5S+l9bDD0QN+plJeYReZcM21RksReTkzVmGNvl974pszpowDsfuTPJhjI7C6YFyp7N0yQDoHHDX
+        o8mlb5ysk4frGlkLhzeFw42IXe8oOS6J9anFUZZ7WZNGuFHcUDg+eThEWbM0in27AyUjg4/huGyH
+        xsHEdiz9VPm/Hj6EnjSIH8+a0fHgpdGeTPb2KI4RXzs2ce/WXkI8Sz4nxUzvbaH3HiOGqClE8sla
+        TalzI8k+stfsDj/DLYbIvtnTnQbAtCqejRs9RxtKjB2hTdCHPm+1N9OMq8TDrFaR/ID/i5Q1HFOq
+        7NUqELNdDVPk0XWXe9eljySp4IO3ZOFHQjPGyr2gHePjYiKmDSpFuENzx84deHIeSXI5mU7UcegV
+        U+bHo1RNF4zyeojfTtcsLnV9i4H/OzG2ZDiw+AAdWkroUglzkBiVEA6z893fHrIVXMP7chMT3JYz
+        M3ON7OGGXTLvsWvNtGI6ZRGfKYCOAGKrMjTtQW7Grw6iSgccAxgcQjFPesI6pmyJ7SkMofh4y5ZA
+        0TeUNggy6BQcV4VTcWJi9guhlmDM5LKchcqPODi4xe7Ai12Jb5qv3J43EtHBLu0idbk7jq1XPGZp
+        gyFialxSTKDbZUkSfie+OXPF50B1TSbO+28TqAx8uSjIk/FuRQON9WbTIjm4E+0HbWQGO5a8ZWeV
+        /IswpTc5Vzw2Bx1Cm/abrHYUHF9nRctN61IGTqNr8RbWFLdEHtBzh65kQr69AmDHNsSSGS1KEeED
+        mnZUNsw9TEqhAIPnPwY6jJyyXb6iyuENgGsZ4rdanlPo8HmE1sEG3ZC7JKXRAlCLAvqdeEodBmm6
+        DG1qemJiPqmZ+EseHp16ftjXKdVDwNRb+sG5gwn0XorEuaP8bZz5MveQTppeZR2+WlrV7Dm0T0oY
+        xKd+MTVdVZ79cgLwW+5Vh6P2s+pVuj4+RXmm/Lnl5dirVvsWeT/75vXFOBsTw/3E5WKaONrwaSz1
+        QbtbGTQt2f3SfW+Mg2U5mDg5oP1XON/au1Bn3/w/2+8njKE+kn3qlSybY8r715TST4i/e20ucwZc
+        BdING3qKTJqksFTj4EpjX4VdiNQ91ewb0l65dPd1/2TXaPDtwtaL6uTLyf8AAAD//wMA10Lu/OsM
+        AAA=
+    headers:
+      CF-RAY:
+      - 937ed42dee2e621f-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:33:48 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=mLRCnpdB3n_6medIZWHnUu8MNRGZsD6riaRhN47PK74-1745930028-1.0.1.1-M2lDM1_V9hNCK0MZrBnFalF3lndC3JkS8zhDOGww_LmOrgdpU9fZLpNZUmyinCQOnlCjDjDYJUECM82ffT1anqBiO1NoDeNp91EPKiK7s.8;
+        path=/; expires=Tue, 29-Apr-25 13:03:48 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=eTrj_ZhCx2XuylS5vYROwUlPrJBwOyrbS2Ki.msl45E-1745930028010-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '10856'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999807'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_bc2d62d8325b2bdd3e98544a66389132
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Guardrail Agent. You
+      are a expert at validating the output of a task. By providing effective feedback
+      if the output is not valid.\nYour personal goal is: Validate the output of the
+      task\n\nTo give my best complete final answer to the task respond using 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!\nIMPORTANT:
+      Your final answer MUST contain all the information requested in the following
+      format: {\n  \"valid\": bool,\n  \"feedback\": str | None\n}\n\nIMPORTANT: Ensure
+      the final output does not include any code block markers like ```json or ```python."},
+      {"role": "user", "content": "\n        Ensure the following task result complies
+      with the given guardrail.\n\n        Task result:\n        Here is a comprehensive
+      list of available books on the First World War:\n\n1. **\"The Sleepwalkers:
+      How Europe Went to War in 1914\" by Christopher Clark**  \n   This book delves
+      into the complex factors that led to the outbreak of the war, offering insights
+      into the political and social dynamics of early 20th century Europe.\n\n2. **\"A
+      World Undone: The Story of the Great War, 1914 to 1918\" by G.J. Meyer**  \n   Meyer''s
+      expansive narrative covers the entire war with a focus on both military strategies
+      and the human experiences endured by soldiers and civilians alike.\n\n3. **\"All
+      Quiet on the Western Front\" by Erich Maria Remarque**  \n   A poignant novel
+      that captures the resilience and trauma experienced by German soldiers during
+      World War I, based on the author''s own experiences.\n\n4. **\"The First World
+      War\" by John Keegan**  \n   Keegan provides a detailed military history of
+      the war, featuring insights on battles, strategies, and the overall impact on
+      global affairs.\n\n5. **\"Goodbye to All That\" by Robert Graves**  \n   This
+      autobiography recounts the author''s experiences as a soldier during the war,
+      offering a personal and critical perspective on the conflicts and the post-war
+      era.\n\n6. **\"Catastrophe 1914: Europe Goes to War\" by Max Hastings**  \n   Hastings
+      chronicles the events leading up to World War I and the early battles, detailing
+      the war''s initial impact on European societies.\n\n7. **\"The War That Ended
+      Peace: The Road to 1914\" by Margaret MacMillan**  \n   MacMillan explores the
+      political and historical factors that contributed to the outbreak of war, emphasizing
+      the decisions made by leaders across Europe.\n\n8. **\"The First World War:
+      A Complete History\" by Martin Gilbert**  \n   This complete history takes readers
+      through the entirety of the war, from its causes to its aftermath, using a wide
+      range of sources.\n\n9. **\"1914: The Year the World Ended\" by Paul Ham**  \n   Ham
+      focuses on the pivotal year of 1914 and the early war''s devastation, analyzing
+      its long-lasting effects on the world.\n\n10. **\"War Horse\" by Michael Morpurgo**  \n    This
+      children''s novel tells the story of a horse and his experiences during the
+      war, highlighting the bond between animals and humans amidst the chaos.\n\nEach
+      of these books offers unique perspectives and rich details about the First World
+      War, making them valuable resources for anyone interested in this pivotal period
+      in history.\n\n        Guardrail:\n        Ensure the authors are from Italy\n        \n        Your
+      task:\n        - Confirm if the Task result complies with the guardrail.\n        -
+      If not, provide clear feedback explaining what is wrong (e.g., by how much it
+      violates the rule, or what specific part fails).\n        - Focus only on identifying
+      issues \u2014 do not propose corrections.\n        - If the Task result complies
+      with the guardrail, saying that is valid\n        "}], "model": "gpt-4o-mini",
+      "stop": ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '3917'
+      content-type:
+      - application/json
+      cookie:
+      - __cf_bm=mLRCnpdB3n_6medIZWHnUu8MNRGZsD6riaRhN47PK74-1745930028-1.0.1.1-M2lDM1_V9hNCK0MZrBnFalF3lndC3JkS8zhDOGww_LmOrgdpU9fZLpNZUmyinCQOnlCjDjDYJUECM82ffT1anqBiO1NoDeNp91EPKiK7s.8;
+        _cfuvid=eTrj_ZhCx2XuylS5vYROwUlPrJBwOyrbS2Ki.msl45E-1745930028010-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFPLbtswELz7KxY8y4HsJHWsW9wiQVq0hzx6aBUIa3IlMaFIlVw5NQL/
+        e0Epiew+gF4EamZnuC8+TwCEViIDIWtk2bRmurqm29X1+/PNzYfL06+fFV8svz2sttuF+XHzUSRR
+        4dYPJPlVdSRd0xpi7exAS0/IFF1ni5PT5XE6m896onGKTJRVLU9P3LTRVk/n6fxkmi6ms7MXde20
+        pCAy+D4BAHjuvzFPq+inyCBNXpGGQsCKRPYWBCC8MxERGIIOjJZFMpLSWSbbp35bu66qOYMrsO4J
+        JFqo9IYAoYr5A9rwRB4gtxfaooHz/j+D59wC5GKDRqtcZFCiCZQMYEmk1igfI56LL84SuBK4JsCO
+        a+cDGB2YFGjbo4zhETyFzjCgJyi9a+CK0WyP4NyYA2VDNraY1BjpuCYP0nWWvaaQQOhkDRjgknyD
+        dpv0BnefEkCrhvPNUS5yu9vviaeyCxjnYjtj9gi01jHGS/tp3L8wu7f+G1e13q3Db1JRaqtDXXjC
+        4GzsdWDXip7dTQDu+zl3B6MTrXdNywW7R+qvWywXg58Y12tk370sgWDHaEb87PRVdeBXKGLUJuxt
+        ipAoa1KjdFwr7JR2e8Rkr+o/s/mb91C5ttX/2I+ElNQyqaL1pLQ8rHgM8xRf37/C3rrcJywC+Y2W
+        VLAmHyehqMTODG9ChG1gaopS24p86/XwMMq2SI+X87P5PF2mYrKb/AIAAP//AwD77a3iJgQAAA==
+    headers:
+      CF-RAY:
+      - 937ed6bd68faa435-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 12:35:23 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '1138'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999072'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_2ba1be014a5974ba354aff564e26516a
+    status:
+      code: 200
+      message: OK
+version: 1

tests/cassettes/test_task_guardrail_process_output.yaml

@@ -0,0 +1,250 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Guardrail Agent. You
+      are a expert at validating the output of a task. By providing effective feedback
+      if the output is not valid.\nYour personal goal is: Validate the output of the
+      task\n\nTo give my best complete final answer to the task respond using 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!\nIMPORTANT:
+      Your final answer MUST contain all the information requested in the following
+      format: {\n  \"valid\": bool,\n  \"feedback\": str | None\n}\n\nIMPORTANT: Ensure
+      the final output does not include any code block markers like ```json or ```python."},
+      {"role": "user", "content": "\n        Ensure the following task result complies
+      with the given guardrail.\n\n        Task result:\n        \n        Lorem Ipsum
+      is simply dummy text of the printing and typesetting industry. Lorem Ipsum has
+      been the industry''s standard dummy text ever\n        \n\n        Guardrail:\n        Ensure
+      the result has less than 10 words\n        \n        Your task:\n        - Confirm
+      if the Task result complies with the guardrail.\n        - If not, provide clear
+      feedback explaining what is wrong (e.g., by how much it violates the rule, or
+      what specific part fails).\n        - Focus only on identifying issues \u2014
+      do not propose corrections.\n        - If the Task result complies with the
+      guardrail, saying that is valid\n        "}], "model": "gpt-4o-mini", "stop":
+      ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '1629'
+      content-type:
+      - application/json
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jFPLbtswELz7KxY824GkxLGtW4KiQB+XBmkRtAqENbmSmFAkQVJ2UsP/
+        HlByLKdNgV4IcGdnOPvgbgLApGA5MN5g4K1Vs+ub718rm324+5z+/CLt1dXD5fU3s1jd3Wx//GbT
+        yDDrB+LhlXXGTWsVBWn0AHNHGCiqpouL+XKVZum8B1ojSEVabcPswsxaqeUsS7KLWbKYpcsDuzGS
+        k2c5/JoAAOz6M/rUgp5YDsn0NdKS91gTy49JAMwZFSMMvZc+oA5sOoLc6EC6t37bmK5uQg6fQJst
+        cNRQyw0BQh39A2q/JQdQ6I9So4Kr/p7DrtAABdugkqJgOVSoPE2HYEUk1sgfY7xgtw1BQP8Ijnyn
+        AsTHUWoP6SVsjRN+CvTEiYTUNYSGoO7QCYdSgZKtDGAqqCiaCA1qSJOBBetnOAicFazQ+9MCHVWd
+        x9hk3Sl1AqDWJmAcUt/a+wOyPzZTmdo6s/Z/UFkltfRN6Qi90bFxPhjLenQ/Abjvh9a9mQOzzrQ2
+        lME8Uv/ceTIf9Ni4KyM6Tw9gMAHVCWt+OX1HrxQUUCp/MnbGkTckRuq4I9gJaU6AyUnVf7t5T3uo
+        XOr6f+RHgHOygURpHQnJ31Y8pjmKX+lfaccu94aZJ7eRnMogycVJCKqwU8OCM//sA7VlJXVNzjo5
+        bHlly+R8lS2zLFklbLKfvAAAAP//AwCHe/Jh8wMAAA==
+    headers:
+      CF-RAY:
+      - 937b20ddf9607def-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 01:46:56 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=nHa2kVJI_yO1RIsmZcEednJ1e9UVy1liv_sjBNtSj7Q-1745891216-1.0.1.1-jUH9kFawVBjnbq8sIL2.MQx.p7JvBZWUhqlkNKRlStWSgQxT0eZMPcgq9TCQoJAjuyNwhqfpK4HuX6x5n8UbQgAb6JrWJEG823e6GpGROEA;
+        path=/; expires=Tue, 29-Apr-25 02:16:56 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=gg2UeahMCOOR8YhitRtzDwENMOnTOuQdyTMVJVHG0Mg-1745891216085-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '896'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999631'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_859221ed1aedb26cc9d335004ccf183e
+    status:
+      code: 200
+      message: OK
+- request:
+    body: '{"messages": [{"role": "system", "content": "You are Guardrail Agent. You
+      are a expert at validating the output of a task. By providing effective feedback
+      if the output is not valid.\nYour personal goal is: Validate the output of the
+      task\n\nTo give my best complete final answer to the task respond using 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!\nIMPORTANT:
+      Your final answer MUST contain all the information requested in the following
+      format: {\n  \"valid\": bool,\n  \"feedback\": str | None\n}\n\nIMPORTANT: Ensure
+      the final output does not include any code block markers like ```json or ```python."},
+      {"role": "user", "content": "\n        Ensure the following task result complies
+      with the given guardrail.\n\n        Task result:\n        \n        Lorem Ipsum
+      is simply dummy text of the printing and typesetting industry. Lorem Ipsum has
+      been the industry''s standard dummy text ever\n        \n\n        Guardrail:\n        Ensure
+      the result has less than 500 words\n        \n        Your task:\n        -
+      Confirm if the Task result complies with the guardrail.\n        - If not, provide
+      clear feedback explaining what is wrong (e.g., by how much it violates the rule,
+      or what specific part fails).\n        - Focus only on identifying issues \u2014
+      do not propose corrections.\n        - If the Task result complies with the
+      guardrail, saying that is valid\n        "}], "model": "gpt-4o-mini", "stop":
+      ["\nObservation:"]}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '1630'
+      content-type:
+      - application/json
+      cookie:
+      - __cf_bm=nHa2kVJI_yO1RIsmZcEednJ1e9UVy1liv_sjBNtSj7Q-1745891216-1.0.1.1-jUH9kFawVBjnbq8sIL2.MQx.p7JvBZWUhqlkNKRlStWSgQxT0eZMPcgq9TCQoJAjuyNwhqfpK4HuX6x5n8UbQgAb6JrWJEG823e6GpGROEA;
+        _cfuvid=gg2UeahMCOOR8YhitRtzDwENMOnTOuQdyTMVJVHG0Mg-1745891216085-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.68.2
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.68.2
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-read-timeout:
+      - '600.0'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.9
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: !!binary |
+        H4sIAAAAAAAAAwAAAP//jJJNb9swDIbv/hWEzvHgfHRpfesOG3opsGE7LYXBSLStRZY0iU43BPnv
+        g5wPu10H7GLAfPhSfEkeMgChlShByBZZdt7kH758e1wzbnfbO6o/f1osV3T/+BO7UNNDIWZJ4bY/
+        SPJF9U66zhti7ewJy0DIlKrO16ub27v5srgZQOcUmSRrPOcrl3fa6nxRLFZ5sc7nt2d167SkKEr4
+        ngEAHIZv6tMq+iVKKGaXSEcxYkOivCYBiOBMigiMUUdGy2I2Qukskx1a/9q6vmm5hAew7hkkWmj0
+        ngChSf0D2vhMAWBjP2qLBu6H/xIOGwuwEXs0Wm1ECRx6mp1iNZHaotylsO2N2djj9PFAdR/RnOEE
+        oLWOMQ1wsP10JserUeMaH9w2vpKKWlsd2yoQRmeTqcjOi4EeM4CnYaD9ixkJH1znuWK3o+G583KG
+        4Vz2ONLF7RmyYzQT1XI5e6NepYhRmzhZiZAoW1KjdNwf9kq7Ccgmrv/u5q3aJ+faNv9TfgRSkmdS
+        lQ+ktHzpeEwLlM78X2nXKQ8Ni0hhryVVrCmkTSiqsTen4xPxd2TqqlrbhoIP+nSBta/SueD7QtWF
+        yI7ZHwAAAP//AwAiLXhqjwMAAA==
+    headers:
+      CF-RAY:
+      - 937b2311ee091b1b-GRU
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Tue, 29 Apr 2025 01:48:26 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '610'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999631'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_c136835c16be6bc1e4d820f239c4b620
+    status:
+      code: 200
+      message: OK
+version: 1

tests/config/tasks.yaml

@@ -6,6 +6,7 @@ research_task:
   expected_output: >
     A list with 10 bullet points of the most relevant information about {topic}
   agent: researcher
+  guardrail: ensure each bullet contains its source
 
 reporting_task:
   description: >

tests/project_test.py

@@ -1,4 +1,5 @@
 from typing import List
+from unittest.mock import patch
 
 import pytest
 
@@ -142,6 +143,15 @@ def test_agent_function_calling_llm():
     ), "agent's function_calling_llm is incorrect"
 
 
+def test_task_guardrail():
+    crew = InternalCrew()
+    research_task = crew.research_task()
+    assert research_task.guardrail == "ensure each bullet contains its source"
+
+    reporting_task = crew.reporting_task()
+    assert reporting_task.guardrail is None
+
+
 @pytest.mark.vcr(filter_headers=["authorization"])
 def test_before_kickoff_modification():
     crew = InternalCrew()

tests/test_task_guardrails.py

@@ -1,11 +1,16 @@
-"""Tests for task guardrails functionality."""
-
-from unittest.mock import Mock
+from unittest.mock import ANY, Mock, patch
 
 import pytest
 
-from crewai.task import Task
+from crewai import Agent, Task
+from crewai.llm import LLM
+from crewai.tasks.task_guardrail import TaskGuardrail
 from crewai.tasks.task_output import TaskOutput
+from crewai.utilities.events import (
+    TaskGuardrailCompletedEvent,
+    TaskGuardrailStartedEvent,
+)
+from crewai.utilities.events.crewai_event_bus import crewai_event_bus
 
 
 def test_task_without_guardrail():
@@ -22,7 +27,7 @@ def test_task_without_guardrail():
     assert result.raw == "test result"
 
 
-def test_task_with_successful_guardrail():
+def test_task_with_successful_guardrail_func():
     """Test that successful guardrail validation passes transformed result."""
 
     def guardrail(result: TaskOutput):
@@ -127,3 +132,138 @@ def test_guardrail_error_in_context():
 
     assert "Task failed guardrail validation" in str(exc_info.value)
     assert "Expected JSON, got string" in str(exc_info.value)
+
+
+@pytest.fixture
+def sample_agent():
+    return Agent(role="Test Agent", goal="Test Goal", backstory="Test Backstory")
+
+
+@pytest.fixture
+def task_output():
+    return TaskOutput(
+        raw="""
+        Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever
+        """,
+        description="Test task",
+        expected_output="Output",
+        agent="Test Agent",
+    )
+
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_task_guardrail_process_output(task_output):
+    guardrail = TaskGuardrail(
+        description="Ensure the result has less than 10 words", llm=LLM(model="gpt-4o")
+    )
+
+    result = guardrail(task_output)
+    assert result[0] is False
+
+    assert "exceeding the guardrail limit of fewer than" in result[1].lower()
+
+    guardrail = TaskGuardrail(
+        description="Ensure the result has less than 500 words", llm=LLM(model="gpt-4o")
+    )
+
+    result = guardrail(task_output)
+    assert result[0] is True
+    assert result[1] == task_output.raw
+
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_guardrail_emits_events(sample_agent):
+    started_guardrail = []
+    completed_guardrail = []
+
+    with crewai_event_bus.scoped_handlers():
+
+        @crewai_event_bus.on(TaskGuardrailStartedEvent)
+        def handle_guardrail_started(source, event):
+            started_guardrail.append(
+                {"guardrail": event.guardrail, "retry_count": event.retry_count}
+            )
+
+        @crewai_event_bus.on(TaskGuardrailCompletedEvent)
+        def handle_guardrail_completed(source, event):
+            completed_guardrail.append(
+                {
+                    "success": event.success,
+                    "result": event.result,
+                    "error": event.error,
+                    "retry_count": event.retry_count,
+                }
+            )
+
+        task = Task(
+            description="Gather information about available books on the First World War",
+            agent=sample_agent,
+            expected_output="A list of available books on the First World War",
+            guardrail="Ensure the authors are from Italy",
+        )
+
+        result = task.execute_sync(agent=sample_agent)
+
+        def custom_guardrail(result: TaskOutput):
+            return (True, "good result from callable function")
+
+        task = Task(
+            description="Test task",
+            expected_output="Output",
+            guardrail=custom_guardrail,
+        )
+
+        task.execute_sync(agent=sample_agent)
+
+        expected_started_events = [
+            {"guardrail": "Ensure the authors are from Italy", "retry_count": 0},
+            {"guardrail": "Ensure the authors are from Italy", "retry_count": 1},
+            {
+                "guardrail": """def custom_guardrail(result: TaskOutput):
+            return (True, "good result from callable function")""",
+                "retry_count": 0,
+            },
+        ]
+
+        expected_completed_events = [
+            {
+                "success": False,
+                "result": None,
+                "error": "The task result does not comply with the guardrail because none of "
+                "the listed authors are from Italy. All authors mentioned are from "
+                "different countries, including Germany, the UK, the USA, and others, "
+                "which violates the requirement that authors must be Italian.",
+                "retry_count": 0,
+            },
+            {"success": True, "result": result.raw, "error": None, "retry_count": 1},
+            {
+                "success": True,
+                "result": "good result from callable function",
+                "error": None,
+                "retry_count": 0,
+            },
+        ]
+        assert started_guardrail == expected_started_events
+        assert completed_guardrail == expected_completed_events
+
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_guardrail_when_an_error_occurs(sample_agent, task_output):
+    with (
+        patch(
+            "crewai.Agent.kickoff",
+            side_effect=Exception("Unexpected error"),
+        ),
+        pytest.raises(
+            Exception,
+            match="Error while validating the task output: Unexpected error",
+        ),
+    ):
+        task = Task(
+            description="Gather information about available books on the First World War",
+            agent=sample_agent,
+            expected_output="A list of available books on the First World War",
+            guardrail="Ensure the authors are from Italy",
+            max_retries=0,
+        )
+        task.execute_sync(agent=sample_agent)