bugfixes: transaction parameter collector, generic formatting messages, guardrails, prompts. Testing contas

2026-09-07 18:23:46 +00:00 · 2026-08-20 00:23:50 -03:00
parent f9c66b4792
commit 762f6fb024
231 changed files with 3999 additions and 141 deletions
--- a/tests/test_transactional_tool_flow.py
+++ b/tests/test_transactional_tool_flow.py
@@ -88,3 +88,202 @@ async def test_transaction_waits_then_executes_after_confirmation():
    assert runtime.calls[-1][0] == "solicitar_devolucao"
    assert runtime.calls[-1][1]["confirmed"] is True
    assert second[-1]["ok"] is True
+
+class _ContestPolicyRouter:
+    def __init__(self):
+        from types import SimpleNamespace
+        self.registry = SimpleNamespace(
+            tools={"contestar_cobranca": object()},
+            get_tool=lambda name: SimpleNamespace(selection_keywords=["contestar", "não contratei", "nao contratei"]),
+        )
+
+    def resolve_execution_policy(self, tool_name, arguments=None):
+        return {
+            "operation_type": "transactional",
+            "require_confirmation": True,
+            "requires": ["subject", "valor"],
+            "policy_source": "test",
+        }
+
+    def validate_execution_policy(self, tool_name, arguments=None):
+        policy = self.resolve_execution_policy(tool_name, arguments)
+        return True, None, policy
+
+
+class _ContestRuntime(AgentRuntimeMixin):
+    def __init__(self):
+        self.tool_router = _ContestPolicyRouter()
+        self.calls = []
+
+    async def _call_mcp_tool(self, tool_name, arguments, state):
+        self.calls.append((tool_name, dict(arguments)))
+        return {"ok": True, "tool_name": tool_name, "result": {"status": "OPENED"}}
+
+
+@pytest.mark.asyncio
+async def test_collecting_parameters_does_not_replace_collected_subject_with_stale_context():
+    """A later value-only turn must not replace an already collected subject.
+
+    Regression reproduced from Contas: subject was collected as TIM CTRL, while
+    context/tool_arguments still exposed TIM Fashion from another transaction.
+    When the user supplied only R$ 71,99, the stale context used to overwrite
+    the collected subject before confirmation.
+    """
+    runtime = _ContestRuntime()
+    state = {
+        "user_text": "R$ 71,99",
+        "sanitized_input": "R$ 71,99",
+        "route": "contestacao_agent",
+        "intent": "state:COLLECTING_CONTESTACAO_PARAMETERS",
+        "transaction_status": "COLLECTING_PARAMETERS",
+        "selected_tool_call": {
+            "tool_name": "contestar_cobranca",
+            "arguments": {
+                "subject": "TIM CTRL Redes Sociais 8.0",
+                "motivo": "não contratei",
+            },
+        },
+        # Simulates stale contextual arguments left by another action/session turn.
+        "context": {
+            "tool_arguments": {
+                "subject": "TIM Fashion Mensal",
+                "valor": 71.99,
+            }
+        },
+    }
+
+    result = await runtime.execute_tools_for_intent(state, tools=[])
+
+    assert result[-1]["transaction_status"] == "AWAITING_CONFIRMATION"
+    assert state["pending_tool_call"]["arguments"]["subject"] == "TIM CTRL Redes Sociais 8.0"
+    assert state["pending_tool_call"]["arguments"]["valor"] == 71.99
+    assert state["pending_tool_call"]["arguments"]["motivo"] == "não contratei"
+    assert state["pending_tool_call"]["arguments"]["query"] == "R$ 71,99"
+    assert runtime.calls == []
+
+class _InitialContestLLM:
+    async def ainvoke(self, messages, **kwargs):
+        prompt = messages[0]["content"]
+        if "Campo: subject" in prompt:
+            return {"content": '{"subject": "TIM CTRL Redes Sociais 8.0"}'}
+        if "Campo: valor" in prompt:
+            return {"content": '{"valor": null}'}
+        if "Campo: motivo" in prompt:
+            return {"content": '{"motivo": "não contratei"}'}
+        return {"content": '{}'}
+
+
+class _InitialContestRouter(_ContestPolicyRouter):
+    def resolve_execution_policy(self, tool_name, arguments=None):
+        return {
+            "operation_type": "transactional",
+            "require_confirmation": True,
+            "requires": ["subject"],
+            "policy_source": "test",
+        }
+
+    def parameter_extract_rules(self, tool_name):
+        return {
+            "subject": {"from": "message", "strategy": "llm", "type": "string", "description": "item"},
+            "valor": {"from": "message", "strategy": "llm", "type": "number", "description": "valor"},
+            "motivo": {"from": "message", "strategy": "llm", "type": "string", "description": "motivo"},
+        }
+
+
+class _InitialContestRuntime(AgentRuntimeMixin):
+    def __init__(self):
+        self.tool_router = _InitialContestRouter()
+        self.llm = _InitialContestLLM()
+        self.calls = []
+
+    async def _call_mcp_tool(self, tool_name, arguments, state):
+        self.calls.append((tool_name, dict(arguments)))
+        return {"ok": True, "tool_name": tool_name, "result": {"status": "OPENED"}}
+
+
+@pytest.mark.asyncio
+async def test_new_contestation_does_not_inherit_subject_or_value_from_previous_transaction():
+    runtime = _InitialContestRuntime()
+    state = {
+        "user_text": "nao contratei TIM CTRL Redes Sociais 8.0",
+        "sanitized_input": "nao contratei TIM CTRL Redes Sociais 8.0",
+        "mcp_tools": ["contestar_cobranca"],
+        "route": "contestacao_agent",
+        "intent": "contas_contestation",
+        "context": {
+            "tool_arguments": {
+                "subject": "TIM Fashion Mensal",
+                "valor": 10.0,
+                "motivo": "contestação antiga",
+            }
+        },
+    }
+
+    result = await runtime.execute_tools_for_intent(state)
+    pending = state["pending_tool_call"]["arguments"]
+    assert result[-1]["transaction_status"] == "AWAITING_CONFIRMATION"
+    assert pending["subject"] == "TIM CTRL Redes Sociais 8.0"
+    assert pending["motivo"] == "não contratei"
+    assert "valor" not in pending
+    assert runtime.calls == []
+
+@pytest.mark.asyncio
+async def test_closed_transaction_is_not_operational_context_for_next_turn():
+    runtime = _InitialContestRuntime()
+    state = {
+        "user_text": "nao contratei TIM CTRL Redes Sociais 8.0",
+        "sanitized_input": "nao contratei TIM CTRL Redes Sociais 8.0",
+        "mcp_tools": ["contestar_cobranca"],
+        "route": "contestacao_agent",
+        "intent": "contas_contestation",
+        # Historical/closed transaction must never feed the new one.
+        "transaction_status": "COMPLETED",
+        "selected_tool_call": {
+            "tool_name": "cancelar_vas_avulso",
+            "arguments": {"subject": "TIM Fashion Mensal", "valor": 10.0},
+        },
+        "pending_tool_call": {},
+        "context": {
+            "tool_arguments": {
+                "subject": "TIM Fashion Mensal",
+                "valor": 10.0,
+            }
+        },
+    }
+
+    result = await runtime.execute_tools_for_intent(state)
+
+    assert result[-1]["transaction_status"] == "AWAITING_CONFIRMATION"
+    assert state["active_transaction"]["tool_name"] == "contestar_cobranca"
+    assert state["active_transaction"]["arguments"]["subject"] == "TIM CTRL Redes Sociais 8.0"
+    assert state["pending_tool_call"]["tool_name"] == "contestar_cobranca"
+    assert state["pending_tool_call"]["arguments"]["subject"] == "TIM CTRL Redes Sociais 8.0"
+    assert state["last_transaction"]["tool_name"] == "cancelar_vas_avulso"
+    assert state["last_transaction"]["arguments"]["subject"] == "TIM Fashion Mensal"
+
+
+@pytest.mark.asyncio
+async def test_terminal_confirmation_closes_active_transaction_and_clears_latches():
+    runtime = _Runtime()
+    state = {
+        "user_text": "Quero devolver o pedido 123 porque me arrependi",
+        "sanitized_input": "Quero devolver o pedido 123 porque me arrependi",
+        "mcp_tools": ["consultar_pedido", "solicitar_devolucao"],
+        "route": "support_agent",
+        "intent": "retail_support_exchange_return",
+    }
+    await runtime.execute_tools_for_intent(state)
+    assert state["active_transaction"]["status"] == "AWAITING_CONFIRMATION"
+
+    state["user_text"] = "sim"
+    state["sanitized_input"] = "sim"
+    await runtime.execute_tools_for_intent(state)
+
+    assert state["transaction_status"] == "COMPLETED"
+    assert state["active_transaction"] is None
+    assert state["selected_tool_call"] == {}
+    assert state["pending_tool_call"] == {}
+    assert state["missing_parameters"] == []
+    assert state["next_state"] is None
+    assert state["last_transaction"]["tool_name"] == "solicitar_devolucao"
+    assert state["last_transaction"]["status"] == "COMPLETED"