Description

When using a ConditionalTask with output_pydantic, a custom guardrail, and context from another task that also uses output_pydantic, CrewAI internally calls json.loads() on a Pydantic object instead of its serialized JSON string, causing a Type Error.

Error Unexpected error during model conversion: TypeError: the JSON object must be str, bytes or bytearray, not DrugClassificationOutput. Returning original result.

Crew Execution Failed

Steps to Reproduce

1. Create a Task with output_pydantic set to a Pydantic model.
2. Create a ConditionalTask that references the first task in context, and also has its own output_pydantic, a condition function, and a guardrail function.
3. Add both tasks to a Crew with Process.sequential.
4. Run crew.kickoff().
5. Observe the TypeError: the JSON object must be str, bytes or bytearray, not pydantic error.

Expected behavior

Expected Behavior

The guardrail should receive result.pydantic as a proper Pydantic object and result.raw as a JSON string. The crew should complete successfully.

Actual Behavior

CrewAI internally attempts json.loads() on the DrugClassificationOutput Pydantic object (from the context task) instead of its serialized string, raising: TypeError: the JSON object must be str, bytes or bytearray, not DrugClassificationOutput

Screenshots/Code snippets

from typing import List, Tuple, Any, Optional from pydantic import BaseModel, Field from crewai import Agent, Task, Crew, Process from crewai.tasks.conditional_task import ConditionalTask from crewai.tasks.task_output import TaskOutput

Pydantic Models

class DrugClassificationOutput(BaseModel): company_drugs: List[str] = Field(default_factory=list) other_drugs: List[str] = Field(default_factory=list)

class ValidationOutput(BaseModel): original_classification: bool validated_classification: bool validation_reason: str elements_found: dict = Field(default_factory=dict)

Condition

def is_nr_classified(output) -> bool: return True

Guardrail

def validation_guardrail(result: TaskOutput) -> Tuple[bool, Any]: try: output = result.pydantic return (True, result.raw) except Exception as e: return (False, f"Validation error: {str(e)}")

Agents

classification_agent = Agent( role="Drug Classifier", goal="Classify drugs", backstory="Expert pharmacologist", )

validation_agent = Agent( role="Validator", goal="Validate classifications", backstory="Expert reviewer", )

Tasks

drug_classification_task = Task( description="Classify drugs in: {text}", expected_output="DrugClassificationOutput object", agent=classification_agent, output_pydantic=DrugClassificationOutput,
)

validation_task = ConditionalTask( description="Validate the classification for: {text}", expected_output="ValidationOutput object", condition=is_nr_classified, agent=validation_agent, context=[drug_classification_task],
guardrail=validation_guardrail,
output_pydantic=ValidationOutput,
)

Crew

crew = Crew( agents=[classification_agent, validation_agent], tasks=[drug_classification_task, validation_task], process=Process.sequential, verbose=True, )

result = crew.kickoff(inputs={"text": "Patient received Clindamycin and developed rash."})

Operating System

Windows 11

Python Version

3.12

crewAI Version

CrewAI Version: 1.14.0

crewAI Tools Version

Virtual Environment

Venv

Evidence

<img width="975" height="350" alt="Image" src="https://github.com/user-attachments/assets/0ed963b7-da3e-4c78-b84e-c1521f590a38" />

Possible Solution

None

Additional context

Analysis The issue appears to be in the internal serialization path when:

1. A context task has output_pydantic set
2. A ConditionalTask consumes that context
3. The ConditionalTask also has a guardrail and its own output_pydantic The Pydantic object from the context task is passed directly to json.loads() without calling .model_dump_json() first.