advent259141
52 lines
1.9 KiB
Python
52 lines
1.9 KiB
Python
from typing import Generic
|
|
|
|
from .agent import Agent
|
|
from .run_context import TContext
|
|
from .tool import FunctionTool
|
|
|
|
|
|
class HandoffTool(FunctionTool, Generic[TContext]):
|
|
"""Handoff tool for delegating tasks to another agent."""
|
|
|
|
def __init__(
|
|
self,
|
|
agent: Agent[TContext],
|
|
parameters: dict | None = None,
|
|
tool_description: str | None = None,
|
|
**kwargs,
|
|
):
|
|
self.agent = agent
|
|
|
|
# Avoid passing duplicate `description` to the FunctionTool dataclass.
|
|
# Some call sites (e.g. SubAgentOrchestrator) pass `description` via kwargs
|
|
# to override what the main agent sees, while we also compute a default
|
|
# description here.
|
|
# `tool_description` is the public description shown to the main LLM.
|
|
# Keep a separate kwarg to avoid conflicting with FunctionTool's `description`.
|
|
description = tool_description or self.default_description(agent.name)
|
|
super().__init__(
|
|
name=f"transfer_to_{agent.name}",
|
|
parameters=parameters or self.default_parameters(),
|
|
description=description,
|
|
**kwargs,
|
|
)
|
|
|
|
# Optional provider override for this subagent. When set, the handoff
|
|
# execution will use this chat provider id instead of the global/default.
|
|
self.provider_id: str | None = None
|
|
|
|
def default_parameters(self) -> dict:
|
|
return {
|
|
"type": "object",
|
|
"properties": {
|
|
"input": {
|
|
"type": "string",
|
|
"description": "The input to be handed off to another agent. This should be a clear and concise request or task.",
|
|
},
|
|
},
|
|
}
|
|
|
|
def default_description(self, agent_name: str | None) -> str:
|
|
agent_name = agent_name or "another"
|
|
return f"Delegate tasks to {self.name} agent to handle the request."
|