# task.py:114
class Task(BaseModel):
    # ...几十个字段，但只有两个是必填...

Task 字段虽多，真正必填的只有两个（task.py:146）：

# task.py:146
description: str = Field(description="Description of the actual task.")
expected_output: str = Field(
    description="Clear definition of expected output for the task.")

两个关系字段决定了这张单子怎么融入团队（task.py:157）：

# task.py:157
agent: Annotated[
    BaseAgent | None,
    BeforeValidator(_resolve_agent),
] = Field(description="Agent responsible for execution the task.", default=None)

# task.py:161
context: list[Task] | None | _NotSpecified = Field(
    description="Other tasks that will have their output used as context for this task.",
    default=NOT_SPECIFIED)

一批字段决定"结果长成什么样、存到哪"（task.py:165 起）：

# task.py:165
async_execution: bool | None = Field(default=False, ...)   # 是否异步执行（不阻塞后面）
output_json: type[BaseModel] | None = Field(default=None,..)# 结果解析成 JSON（给个模型类当模板）
output_pydantic: type[BaseModel] | None = ...              # 结果解析成 Pydantic 对象
response_model: type[BaseModel] | None = ...               # 用 provider 原生结构化输出
output_file: str | None = Field(default=None, ...)         # 结果写到文件
create_directory: bool | None = Field(default=True, ...)   # 文件目录不存在就自动建
output: TaskOutput | None = Field(default=None, ...)        # ★执行后回填的结果对象
tools: list[BaseTool] | None = Field(default_factory=list,.)# 本任务限定可用的工具

Task 不只是数据，它带执行方法。execute_sync 是同步执行入口（task.py:572），真正逻辑在 _execute_core（task.py:762）：

# task.py:572
def execute_sync(self, agent=None, context=None, tools=None) -> TaskOutput:
    """Execute the task synchronously."""
    self.start_time = datetime.datetime.now()
    return self._execute_core(agent, context, tools)

# task.py:762 —— 核心执行（裁剪）
def _execute_core(self, agent, context, tools) -> TaskOutput:
    agent = agent or self.agent
    self.agent = agent
    if not agent:
        raise Exception(f"The task '{self.description}' has no agent assigned, ...")  # ★没人干 → 报错
    self.prompt_context = context
    tools = tools or self.tools or []
    self.processed_by_agents.add(agent.role)
    crewai_event_bus.emit(self, TaskStartedEvent(context=context, task=self))  # 发"任务开始"事件
    result = agent.execute_task(task=self, context=context, tools=tools)        # ★把活真正交给 agent
    self._post_agent_execution(agent)
    # ...根据 output_pydantic/output_json 决定 raw/pydantic/json_output（见 L06）...

agent 返回结果后，_execute_core 根据你设的输出形态字段，决定怎么装配（task.py:798）：

# task.py:798
if isinstance(result, BaseModel):                 # agent 直接返回了结构化对象
    raw = result.model_dump_json()
    if self.output_pydantic:
        pydantic_output = result; json_output = None
    elif self.output_json:
        pydantic_output = None; json_output = result.model_dump()
    else:
        pydantic_output = None; json_output = None
elif not self._guardrails and not self._guardrail:
    raw = result
    pydantic_output, json_output = self._export_output(result)  # 从文本里尝试解析结构化
else:
    raw = result; pydantic_output, json_output = None, None

# task.py:816
task_output = TaskOutput(
    name=self.name or self.description,
    description=self.description,
    expected_output=self.expected_output,
    raw=raw,
    pydantic=pydantic_output,
    json_dict=json_output,
    agent=agent.role,
    output_format=self._get_output_format(),
    messages=agent.last_messages)