234 lines
8.0 KiB
Python
234 lines
8.0 KiB
Python
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from app.extensions.tim_guardrails import TimProactiveOfferRail
|
|
|
|
|
|
class _FailIfCalledLLM:
|
|
async def ainvoke(self, *args, **kwargs):
|
|
raise AssertionError('TIM_AOFERTA LLM must not run for transaction continuation')
|
|
|
|
|
|
class _BlockingLLM:
|
|
def __init__(self):
|
|
self.calls = 0
|
|
|
|
async def ainvoke(self, *args, **kwargs):
|
|
self.calls += 1
|
|
return '{"allowed": false, "reason": "oferta proativa"}'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
@pytest.mark.parametrize('status', ['COLLECTING_PARAMETERS', 'AWAITING_CONFIRMATION'])
|
|
async def test_tim_aoferta_bypasses_native_transaction_continuation_states(status):
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Você confirma o cancelamento do serviço TIM Fashion?',
|
|
{
|
|
'transaction_status': status,
|
|
'guardrail_llm': _FailIfCalledLLM(),
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.code == 'TIM_AOFERTA'
|
|
assert decision.reason == f'continuidade_transacional:{status}'
|
|
assert decision.metadata['mechanism'] == 'deterministic_transaction_bypass'
|
|
assert decision.metadata['transaction_status'] == status
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_detects_awaiting_confirmation_from_mcp_results():
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Você confirma o cancelamento do serviço TIM Fashion?',
|
|
{
|
|
'mcp_results': [
|
|
{
|
|
'tool_name': 'cancelar_vas_avulso',
|
|
'awaiting_confirmation': True,
|
|
'transaction_status': 'AWAITING_CONFIRMATION',
|
|
}
|
|
],
|
|
'guardrail_llm': _FailIfCalledLLM(),
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.metadata['transaction_status'] == 'AWAITING_CONFIRMATION'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_detects_parameter_collection_from_tool_result():
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Para prosseguir, informe valor.',
|
|
{
|
|
'tool_result': [
|
|
{
|
|
'tool_name': 'contestar_cobranca',
|
|
'transaction_status': 'COLLECTING_PARAMETERS',
|
|
}
|
|
],
|
|
'guardrail_llm': _FailIfCalledLLM(),
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.metadata['transaction_status'] == 'COLLECTING_PARAMETERS'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_still_uses_llm_outside_transaction_continuation():
|
|
llm = _BlockingLLM()
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Caso queira, posso cancelar outro serviço.',
|
|
{
|
|
'transaction_status': 'COMPLETED',
|
|
'guardrail_llm': llm,
|
|
},
|
|
)
|
|
|
|
assert llm.calls == 1
|
|
assert decision.allowed is False
|
|
assert decision.reason == 'oferta proativa'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_allows_structurally_authorized_human_handoff_without_calling_llm():
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Vou encaminhar seu atendimento para uma pessoa.',
|
|
{
|
|
'current_route': 'human_handoff',
|
|
'current_intent': 'human_handoff',
|
|
'session_control': 'HUMAN_HANDOFF',
|
|
'human_handoff_requested': True,
|
|
'guardrail_llm': _FailIfCalledLLM(),
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.reason == 'handoff_humano_autorizado'
|
|
assert decision.metadata['mechanism'] == 'deterministic_handoff_bypass'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_does_not_bypass_transfer_phrase_without_structural_handoff():
|
|
llm = _BlockingLLM()
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Vou encaminhar seu atendimento para uma pessoa.',
|
|
{
|
|
'current_route': 'faturas_agent',
|
|
'current_intent': 'contas_invoice_query',
|
|
'guardrail_llm': llm,
|
|
},
|
|
)
|
|
|
|
assert llm.calls == 1
|
|
assert decision.allowed is False
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_allows_terminal_handoff_from_resumed_workflow_without_router_handoff():
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Para continuar com a sua solicitação, aguarde um instante.',
|
|
{
|
|
'current_route': 'faturas_agent',
|
|
'current_intent': 'contas_invoice_explanation',
|
|
'mcp_results': [{
|
|
'tool_name': 'retomar_workflow',
|
|
'result': {
|
|
'status': 'COMPLETED',
|
|
'output': {
|
|
'mensagem': 'Para continuar com a sua solicitação, aguarde um instante.',
|
|
'session_control': 'HUMAN_HANDOFF',
|
|
'human_handoff_requested': True,
|
|
'handoff': True,
|
|
'session_ended': True,
|
|
'terminal_status': 'human_handoff',
|
|
},
|
|
},
|
|
}],
|
|
'guardrail_llm': _FailIfCalledLLM(),
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.reason == 'handoff_humano_autorizado'
|
|
assert decision.metadata['mechanism'] == 'deterministic_handoff_bypass'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_does_not_trust_non_terminal_workflow_handoff_flag():
|
|
llm = _BlockingLLM()
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Vou encaminhar seu atendimento para uma pessoa.',
|
|
{
|
|
'current_route': 'faturas_agent',
|
|
'current_intent': 'contas_invoice_explanation',
|
|
'mcp_results': [{
|
|
'result': {
|
|
'output': {
|
|
'handoff': True,
|
|
},
|
|
},
|
|
}],
|
|
'guardrail_llm': llm,
|
|
},
|
|
)
|
|
|
|
assert llm.calls == 1
|
|
assert decision.allowed is False
|
|
|
|
class _CaptureAllowLLM:
|
|
def __init__(self):
|
|
self.prompt = ''
|
|
|
|
async def ainvoke(self, messages, **kwargs):
|
|
self.prompt = messages[-1]['content']
|
|
return '{"allowed": true, "reason": ""}'
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_tim_aoferta_terminal_fallback_receives_compact_authoritative_transaction_context():
|
|
llm = _CaptureAllowLLM()
|
|
huge = {'noise': 'x' * 50000}
|
|
decision = await TimProactiveOfferRail().evaluate(
|
|
'Não consegui cancelar o Paramount+ por aqui. Use o canal oficial para concluir esse mesmo cancelamento.',
|
|
{
|
|
'transaction_status': 'COMPLETED',
|
|
'current_user_message': 'isso mesmo, pode cancelar',
|
|
'conversation_history': [
|
|
{'role': 'user', 'content': 'quero cancelar Tamboro e Paramount+'},
|
|
{'role': 'assistant', 'content': 'Você confirma o cancelamento de Tamboro e Paramount+?'},
|
|
{'role': 'user', 'content': 'isso mesmo, pode cancelar'},
|
|
],
|
|
'transaction_pre_validation': {
|
|
'requested_arguments': {'subject': 'Tamboro e Paramount+'},
|
|
'resolved_arguments': {'subject': 'Tamboro Mensal, Paramount+', 'items': [{'name': 'Tamboro Mensal'}, {'name': 'Paramount+'}]},
|
|
'effective_tool_name': 'cancelar_vas_avulso',
|
|
},
|
|
'mcp_results': [
|
|
huge,
|
|
{
|
|
'tool_name': 'cancelar_vas_avulso',
|
|
'ok': True,
|
|
'result': {
|
|
'status': 'COMPLETED',
|
|
'output': {
|
|
'results': [
|
|
{'subject': 'Tamboro Mensal', 'success': True},
|
|
{'subject': 'Paramount+', 'success': False, 'reason': 'service_not_found'},
|
|
]
|
|
},
|
|
},
|
|
},
|
|
],
|
|
'guardrail_llm': llm,
|
|
},
|
|
)
|
|
|
|
assert decision.allowed is True
|
|
assert decision.metadata['mechanism'] == 'llm_semantic_compact_transaction_context'
|
|
assert 'Paramount+' in llm.prompt
|
|
assert 'service_not_found' in llm.prompt
|
|
assert 'isso mesmo, pode cancelar' in llm.prompt
|
|
assert len(llm.prompt) < 30000
|