GONATURELLE for AutoGen

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.conditions import TextMentionTermination, MaxMessageTermination
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_ext.models.openai import OpenAIChatCompletionClient
from gonaturelle import EvalSuite, TrajectoryTester
from gonaturelle.integrations.autogen import AutoGenAdapter

from app.tools import lookup_symptoms, escalate_to_clinician

def build_team() -> RoundRobinGroupChat:
    client = OpenAIChatCompletionClient(model='gpt-4.1')
    triage = AssistantAgent('triage', model_client=client, tools=[lookup_symptoms, escalate_to_clinician])
    reviewer = AssistantAgent('reviewer', model_client=client, system_message='Reply APPROVED only if red flags were escalated.')
    stop = TextMentionTermination('APPROVED') | MaxMessageTermination(12)
    return RoundRobinGroupChat([triage, reviewer], termination_condition=stop)

agent = AutoGenAdapter(team_factory=build_team, agent='triage-team')

def test_triage_team():
    results = EvalSuite.load('healthcare/triage-escalation').evaluate(agent=agent, max_concurrency=5)
    t = TrajectoryTester(results)
    t.assert_called('escalate_to_clinician', when={'tags': 'red-flag'})
    t.assert_stop_reason_not('MaxMessageTermination')