类的定义与定位（lite_agent.py:187）：

# lite_agent.py:187
class LiteAgent(FlowTrackable, BaseModel):
    """...
    Use ``Agent().kickoff(messages)`` instead, which provides the same ...
    """

LiteAgent 的字段（lite_agent.py:212），和 Day 07 的重型 Agent 对照着看：

# lite_agent.py:212
id: UUID4 = Field(default_factory=uuid.uuid4, frozen=True)
role: str = Field(description="Role of the agent")            # 身份三件套还在
goal: str = Field(description="Goal of the agent")
backstory: str = Field(description="Backstory of the agent")
llm: str | BaseLLM | Any | None = Field(default=None, ...)   # 接大脑还在
tools: list[BaseTool] = Field(default_factory=list, ...)     # 工具还在
max_iterations: int = Field(default=15, ...)                 # 循环上限（注意默认 15，不是 25！）
max_execution_time: int | None = Field(default=None, ...)
respect_context_window: bool = Field(default=True, ...)      # 上下文压缩还在
use_stop_words: bool = Field(default=True, ...)
response_format: type[BaseModel] | None = Field(default=None, ...)  # 结构化输出
verbose: bool = Field(default=False, ...)
guardrail: Annotated[...] = Field(...)                       # 护栏还在
guardrail_max_retries: int = Field(default=3, ...)
memory: bool | Any | None = Field(default=None, ...)         # 可选记忆

入口 kickoff（lite_agent.py:477）——注意它直接吃字符串或消息列表：

# lite_agent.py:477
def kickoff(self, messages: str | list[LLMMessage],
            response_format: type[BaseModel] | None = None,
            input_files: dict[str, FileInput] | None = None) -> LiteAgentOutput:
    if self._memory is not None:              # 可选：把记忆能力包成工具挂上去
        ...
        self._parsed_tools = self._parsed_tools + parse_tools(memory_tools)
    agent_info = {"id": self.id, "role": self.role, "goal": self.goal, ...}
    try:
        self._iterations = 0                  # ① 循环计数归零
        self.tools_results = []
        self._messages = self._format_messages(messages, response_format=..., input_files=...)  # ② 组装消息
        self._inject_memory_context()         # ③ 注入记忆上下文（若有）
        return self._execute_core(agent_info=agent_info, response_format=response_format)  # ④ 进核心
    except Exception as e:
        ...
        crewai_event_bus.emit(self, event=LiteAgentExecutionErrorEvent(...))
        raise e

核心循环 _invoke_loop（lite_agent.py:860），和 Day 08 几乎是双胞胎：

# lite_agent.py:860
def _invoke_loop(self, response_model=None) -> AgentFinish:
    formatted_answer: AgentAction | AgentFinish | None = None
    while not isinstance(formatted_answer, AgentFinish):        # ★同样的循环条件
        try:
            if has_reached_max_iterations(self._iterations, self.max_iterations):
                formatted_answer = handle_max_iterations_exceeded(...)   # 同样的刹车
            enforce_rpm_limit(self.request_within_rpm_limit)            # 同样的限流
            answer = get_llm_response(llm=..., messages=self._messages, ...)  # 同样的问 LLM
            if isinstance(answer, BaseModel):
                formatted_answer = AgentFinish(thought="", output=answer, text=answer.model_dump_json())
                break
            formatted_answer = process_llm_response(cast(str, answer), self.use_stop_words)  # 同样的解析(D09)
            if isinstance(formatted_answer, AgentAction):
                tool_result = execute_tool_and_check_finality(   # 同样的执行工具
                    agent_action=formatted_answer, tools=self._parsed_tools,
                    agent_key=self.key, agent_role=self.role, agent=self.original_agent, crew=None)
                formatted_answer = handle_agent_action_core(   # 同样的把结果写回
                    formatted_answer=formatted_answer, tool_result=tool_result, show_logs=self._show_logs)
            self._append_message(formatted_answer.text, role="assistant")
        except OutputParserError as e:
            formatted_answer = handle_output_parser_exception(e=e, ...)   # 同样的自愈
    ...

_execute_core（lite_agent.py:619）在循环之后做结果封装和护栏校验：

# lite_agent.py:619
def _execute_core(self, agent_info, response_format=None) -> LiteAgentOutput:
    ...
    agent_finish = self._invoke_loop(response_model=active_response_format)  # ① 跑循环
    ...
    # ② 尽力把输出转成 response_format 指定的 Pydantic 对象
    if isinstance(agent_finish.output, BaseModel):
        formatted_result = agent_finish.output
    elif active_response_format:
        try:
            formatted_result = active_response_format.model_validate_json(str(agent_finish.output))
        except ValidationError:
            ... # 退而用 Converter 让 LLM 再规整一次
    usage_metrics = self.llm.get_token_usage_summary() if isinstance(self.llm, BaseLLM) else ...
    output = LiteAgentOutput(raw=raw_output, pydantic=formatted_result,   # ③ 打包统一输出
                             agent_role=self.role, usage_metrics=..., messages=self._messages)
    if self._guardrail is not None:                                       # ④ 护栏校验
        guardrail_result = process_guardrail(output=output, guardrail=self._guardrail, ...)
        if not guardrail_result.success:
            if self._guardrail_retry_count >= self.guardrail_max_retries:
                raise Exception(f"Agent's guardrail failed validation after ... retries.")  # 重试用尽→抛
            self._guardrail_retry_count += 1
            self._messages.append({"role": "user", "content": guardrail_result.error or ...})  # 把失败原因喂回
            return self._execute_core(agent_info=agent_info)              # ★递归重跑一遍
    ...
    return output

输出统一封装在 LiteAgentOutput（lite_agent_output.py:30）：

# lite_agent_output.py:30
class LiteAgentOutput(BaseModel):
    model_config = {"arbitrary_types_allowed": True}
    raw: str = Field(description="Raw output of the agent", default="")          # 原始文本
    pydantic: BaseModel | None = Field(description="Pydantic output ...", default=None)  # 结构化对象
    agent_role: str = Field(description="Role of the agent that produced this output")
    usage_metrics: dict[str, Any] | None = Field(default=None, ...)              # token 用量
    messages: list[LLMMessage] = Field(default_factory=list, ...)                # 完整对话
    plan: str | None = Field(default=None, ...)                                  # 若规划过：计划
    todos: list[TodoExecutionResult] = Field(default_factory=list, ...)          # 若规划过：各步结果
    replan_count: int = Field(default=0, ...)

# lite_agent_output.py:104
def __str__(self) -> str:
    return self.raw                            # print(output) 直接显示原始文本