Item 23: Use ChatMiddleware to Transform Individual Messages¶
ChatMiddleware 作用于单个 Chat 消息级别，在消息进入或离开时进行转换、验证、过滤。
消息级别心智模型¶
基础用法¶
from agent_framework.middleware import ChatMiddleware
class InputFilterMiddleware(ChatMiddleware):
"""输入过滤：敏感词替换"""
async def process_input(self, message: ChatMessage):
message.content = message.content.replace("密码", "***")
return message
async def process_output(self, message: ChatMessage):
return message
agent = Agent(
client=client,
middleware=[InputFilterMiddleware()],
)
消息转换链¶
class ChainMiddleware(ChatMiddleware):
def __init__(self, middlewares: list[ChatMiddleware]):
self.middlewares = middlewares
async def process_input(self, message: ChatMessage):
for mw in self.middlewares:
message = await mw.process_input(message)
return message
常见模式¶
| 模式 | 用途 |
|---|---|
| Input Filter | 敏感词过滤、输入验证 |
| Output Transform | 响应改写、格式调整 |
| Audit Log | 记录对话内容 |
| Content Warning | 添加警告信息 |
Things to Remember¶
- ChatMiddleware 作用于单条消息
- 输入过滤在消息进入前处理
- 输出转换在消息发出前处理
- 可以链式组合多个中间件