回看 context 字段声明（task.py:161），它的类型和默认值都很讲究：

# 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,
)

类型是 list[Task] | None | _NotSpecified——三选一，对应三种截然不同的语义：

NOT_SPECIFIED 是一个专门的哨兵对象（sentinel），定义在 utilities/constants.py（utilities/constants.py:46）：

# utilities/constants.py:46
class _NotSpecified:
    """A sentinel type for a value that has not been specified.

    - TODO: Consider moving this class and NOT_SPECIFIED to types.py
    """
    def __repr__(self) -> str:
        return "NOT_SPECIFIED"
    ...
# 之后（:77 附近）
NOT_SPECIFIED: Final[...] = _NotSpecified()
#   "Unlike `None`, which might be a valid value from the user, `NOT_SPECIFIED`
#    ..."（源码注释点明了它存在的理由）

执行任务前，Crew 用 _get_context 算出要喂给这个任务的上下文字符串（crew.py:1827）——它正好按三态分三支：

# crew.py:1827
@staticmethod
def _get_context(task: Task, task_outputs: list[TaskOutput]) -> str:
    if not task.context:                        # ① None 或 空列表 → 无上下文
        return ""
    return (
        aggregate_raw_outputs_from_task_outputs(task_outputs)   # ② NOT_SPECIFIED → 用"目前为止所有产出"
        if task.context is NOT_SPECIFIED
        else aggregate_raw_outputs_from_tasks(task.context)     # ③ 指定列表 → 只用这些任务的产出
    )

"汇总"具体怎么做？看 utilities/formatter.py（utilities/formatter.py:16、utilities/formatter.py:29）：

# utilities/formatter.py:13
DIVIDERS: Final[str] = "\n\n----------\n\n"

# utilities/formatter.py:16
def aggregate_raw_outputs_from_task_outputs(task_outputs: list[TaskOutput]) -> str:
    """Generate string context from the task outputs."""
    return DIVIDERS.join(output.raw for output in task_outputs)   # ★把各 raw 用分隔线连起来

# utilities/formatter.py:29
def aggregate_raw_outputs_from_tasks(tasks: list[Task] | _NotSpecified) -> str:
    task_outputs = (
        [task.output for task in tasks if task.output is not None]  # 取每个任务已回填的 output
        if isinstance(tasks, list)
        else []
    )
    return aggregate_raw_outputs_from_task_outputs(task_outputs)

依赖关系必须是"向后看"的——你只能依赖排在你前面的任务。Crew 有个校验器专门拦截"依赖未来任务"（crew.py:838）：

# crew.py:838
@model_validator(mode="after")
def validate_context_no_future_tasks(self) -> Self:
    """Validates that a task's context does not include future tasks."""
    task_indices = {id(task): i for i, task in enumerate(self.tasks)}   # 每个任务的位置
    for task in self.tasks:
        if isinstance(task.context, list):
            for context_task in task.context:
                if id(context_task) not in task_indices:
                    continue
                if task_indices[id(context_task)] > task_indices[id(task)]:  # 依赖者在后
                    raise ValueError(
                        f"Task '{task.description}' has a context dependency "
                        f"on a future task '{context_task.description}', "
                        f"which is not allowed.")
    return self

Crew 常要深拷贝任务（比如 kickoff_for_each 多份输入各跑一遍）。复制时有个难题：拷贝出来的任务，它的 context 应该指向拷贝后的上游任务，而不是原来的。copy() 用 D13 讲的 key（内容指纹）来重连（task.py:1095）：

# task.py:1095（copy 内部）
cloned_context = (
    self.context
    if self.context is NOT_SPECIFIED                          # 默认态：原样保留
    else [task_mapping[context_task.key] for context_task in self.context]  # ★按 key 查新任务
    if isinstance(self.context, list)
    else None                                                  # None：还是 None
)

✋ 10 分钟动手

P=lib/crewai/src/crewai
# 1. context 字段 + 哨兵定义
sed -n '161,164p' $P/task.py
sed -n '46,82p' $P/utilities/constants.py

# 2. _get_context 三态分流
sed -n '1827,1835p' $P/crew.py

# 3. 拼接实现
sed -n '13,46p' $P/utilities/formatter.py

# 4. 未来任务校验 + copy 重连
sed -n '838,854p' $P/crew.py
sed -n '1095,1101p' $P/task.py