Projeto do Agent Contas ORACLE

This commit is contained in:
2026-08-19 09:35:50 -03:00
commit 950a2bcd33
1366 changed files with 177217 additions and 0 deletions

View File

@@ -0,0 +1,23 @@
import pytest
from types import SimpleNamespace
from app.agents.runtime import AgentRuntimeMixin
class DummyAgent(AgentRuntimeMixin):
def __init__(self):
self.settings = SimpleNamespace(CACHE_TTL_SECONDS=10)
self.calls = 0
self.cache = None
self.rag_service = None
self.telemetry = None
class LLM:
async def ainvoke(inner, messages):
self.calls += 1
return 'ok'
self.llm = LLM()
@pytest.mark.asyncio
async def test_agent_runtime_without_cache_invokes_llm():
agent = DummyAgent()
answer = await agent._invoke_llm_cached({'user_text':'oi'}, 'dummy', [{'role':'user','content':'oi'}])
assert answer == 'ok'
assert agent.calls == 1

View File

@@ -0,0 +1,50 @@
from __future__ import annotations
import base64
import hashlib
from fastapi import FastAPI
from fastapi.testclient import TestClient
from agent_framework.security.authentication import ApiKeyAuthenticationProvider, BasicAuthenticationProvider
from agent_framework.security.middleware import AuthenticationMiddleware
def _basic(value: str) -> str:
return "Basic " + base64.b64encode(value.encode()).decode()
def test_basic_authentication_protects_endpoint_and_keeps_health_public():
app = FastAPI()
app.add_middleware(
AuthenticationMiddleware,
provider=BasicAuthenticationProvider("tia", "sha256:" + hashlib.sha256(b"secret").hexdigest()),
public_paths=["/health"],
)
@app.get("/health")
async def health():
return {"status": "ok"}
@app.get("/protected")
async def protected():
return {"status": "protected"}
client = TestClient(app)
assert client.get("/health").status_code == 200
assert client.get("/protected").status_code == 401
assert client.get("/protected", headers={"Authorization": _basic("tia:wrong")}).status_code == 401
assert client.get("/protected", headers={"Authorization": _basic("tia:secret")}).status_code == 200
def test_api_key_authentication():
app = FastAPI()
app.add_middleware(AuthenticationMiddleware, provider=ApiKeyAuthenticationProvider("plain:key-123"))
@app.get("/protected")
async def protected():
return {"status": "ok"}
client = TestClient(app)
assert client.get("/protected").status_code == 401
assert client.get("/protected", headers={"x-api-key": "key-123"}).status_code == 200

View File

@@ -0,0 +1,69 @@
from __future__ import annotations
import base64
from fastapi import FastAPI, Request
from fastapi.testclient import TestClient
from agent_framework.security import (
AuthenticationPolicy,
BasicAuthenticationProvider,
NoAuthenticationProvider,
PolicyAuthenticationMiddleware,
)
def _basic(client_id: str, secret: str) -> str:
value = base64.b64encode(f"{client_id}:{secret}".encode()).decode()
return f"Basic {value}"
def test_policy_middleware_public_protected_and_default_deny():
app = FastAPI()
policies = [
AuthenticationPolicy("public", NoAuthenticationProvider(), paths=("/health",)),
AuthenticationPolicy(
"messages",
BasicAuthenticationProvider("tia", "plain:secret"),
paths=("/gateway/*",),
),
]
app.add_middleware(PolicyAuthenticationMiddleware, policies=policies)
@app.get("/health")
async def health():
return {"ok": True}
@app.get("/gateway/message")
async def message(request: Request):
return {"subject": request.state.auth_principal.subject}
@app.get("/unknown")
async def unknown():
return {"unexpected": True}
client = TestClient(app)
assert client.get("/health").status_code == 200
assert client.get("/gateway/message").status_code == 401
authenticated = client.get("/gateway/message", headers={"Authorization": _basic("tia", "secret")})
assert authenticated.status_code == 200
assert authenticated.json()["subject"] == "tia"
assert client.get("/unknown").status_code == 401
def test_policy_method_filter():
app = FastAPI()
policies = [AuthenticationPolicy("post-only", NoAuthenticationProvider(), paths=("/resource",), methods=frozenset({"POST"}))]
app.add_middleware(PolicyAuthenticationMiddleware, policies=policies)
@app.get("/resource")
async def get_resource():
return {"method": "GET"}
@app.post("/resource")
async def post_resource():
return {"method": "POST"}
client = TestClient(app)
assert client.post("/resource").status_code == 200
assert client.get("/resource").status_code == 401

View File

@@ -0,0 +1,19 @@
import asyncio
import pytest
from agent_framework.cache.cache import InMemoryCache, DistributedCache
@pytest.mark.asyncio
async def test_in_memory_cache_ttl_expires():
cache = InMemoryCache()
await cache.set('k', {'v': 1}, ttl_seconds=1)
assert await cache.get('k') == {'v': 1}
await asyncio.sleep(1.05)
assert await cache.get('k') is None
@pytest.mark.asyncio
async def test_distributed_cache_promotes_l2_to_l1():
l1, l2 = InMemoryCache(), InMemoryCache()
cache = DistributedCache(l1, l2)
await l2.set('x', 'from-l2')
assert await cache.get('x') == 'from-l2'
assert await l1.get('x') == 'from-l2'

View File

@@ -0,0 +1,12 @@
import pytest
from agent_framework.cache.cache import DistributedCache, InMemoryCache
@pytest.mark.asyncio
async def test_distributed_cache_populates_l1_from_l2():
l1 = InMemoryCache(); l2 = InMemoryCache()
await l2.set("k", {"v": 1})
cache = DistributedCache(l1, l2)
assert await cache.get("k") == {"v": 1}
await l2.delete("k")
assert await cache.get("k") == {"v": 1}

View File

@@ -0,0 +1,5 @@
def test_core_imports():
import agent_framework.cache.cache
import agent_framework.rag.rag_service
import agent_framework.checkpoints.langgraph_saver
import agent_framework.observability.telemetry

View File

@@ -0,0 +1,14 @@
import pytest
from types import SimpleNamespace
from agent_framework.checkpoints.langgraph_saver import RepositoryCheckpointSaver
@pytest.mark.asyncio
async def test_repository_checkpoint_saver_put_get(tmp_path):
settings = SimpleNamespace(CHECKPOINT_REPOSITORY_PROVIDER='sqlite', SQLITE_DB_PATH=str(tmp_path/'db.sqlite'))
saver = RepositoryCheckpointSaver(settings)
config = {'configurable': {'thread_id': 't1'}}
next_config = await saver.aput(config, {'id': 'cp1', 'channel_values': {'x': 1}}, {'source': 'test'}, {})
assert next_config['configurable']['checkpoint_id'] == 'cp1'
tup = await saver.aget_tuple(config)
checkpoint = tup.checkpoint if hasattr(tup, 'checkpoint') else tup['checkpoint']
assert checkpoint['id'] == 'cp1'

View File

@@ -0,0 +1,29 @@
import pytest
from agent_framework.observability.langgraph_telemetry import LangGraphDeepTelemetry
class FakeTelemetry:
def __init__(self): self.events=[]
async def event(self, name, payload=None, kind='event'):
self.events.append((name, payload or {}, kind))
def span(self, name, **attrs):
class CM:
async def __aenter__(self_inner): return None
async def __aexit__(self_inner, exc_type, exc, tb): return False
return CM()
@pytest.mark.asyncio
async def test_langgraph_node_emits_started_completed():
telemetry = FakeTelemetry()
tracer = LangGraphDeepTelemetry(telemetry)
async with tracer.node('router', {'session_id': 's1'}):
pass
names = [e[0] for e in telemetry.events]
assert 'langgraph.node.started' in names
assert 'langgraph.node.completed' in names
@pytest.mark.asyncio
async def test_langgraph_edge_event():
telemetry = FakeTelemetry()
tracer = LangGraphDeepTelemetry(telemetry)
await tracer.edge('routing', 'billing', {'session_id': 's1'}, {'confidence': 0.9})
assert telemetry.events[0][0] == 'langgraph.edge.selected'

View File

@@ -0,0 +1,44 @@
from types import SimpleNamespace
from agent_framework.memory.long_term_store import (
InMemoryLongTermMemoryStore,
OracleAutonomousLongTermMemoryStore,
SQLiteLongTermMemoryStore,
create_long_term_memory_store,
)
def settings(provider: str):
return SimpleNamespace(
LONG_TERM_MEMORY_PROVIDER=provider,
LONG_TERM_MEMORY_SQLITE_PATH=":memory:",
LONG_TERM_MEMORY_TABLE="agentfw_long_term_memory",
LONG_TERM_MEMORY_ORACLE_TABLE=None,
ADB_USER="user",
ADB_PASSWORD="password",
ADB_DSN="service_high",
ADB_WALLET_LOCATION=None,
ADB_WALLET_PASSWORD=None,
ADB_TABLE_PREFIX="AGENTFW",
)
def test_factory_memory():
assert isinstance(create_long_term_memory_store(settings("memory")), InMemoryLongTermMemoryStore)
def test_factory_sqlite():
assert isinstance(create_long_term_memory_store(settings("sqlite")), SQLiteLongTermMemoryStore)
def test_factory_autonomous():
store = create_long_term_memory_store(settings("autonomous"))
assert isinstance(store, OracleAutonomousLongTermMemoryStore)
assert store.table == "AGENTFW_LONG_TERM_MEMORY"
def test_factory_oracle_alias():
assert isinstance(
create_long_term_memory_store(settings("oracle")),
OracleAutonomousLongTermMemoryStore,
)

View File

@@ -0,0 +1,60 @@
from __future__ import annotations
import sys
import types
import pytest
from agent_framework.analytics.providers.pubsub import PubSubAnalyticsPublisher
class _FakeFuture:
def result(self, timeout: float | None = None) -> str:
return "fake-message-id"
class _FakePublisherClient:
def __init__(self) -> None:
self.calls: list[tuple[str, bytes, dict[str, str]]] = []
def publish(self, topic_path: str, *, data: bytes, **kwargs: str) -> _FakeFuture:
self.calls.append((topic_path, data, kwargs))
return _FakeFuture()
@pytest.fixture
def pubsub_publisher(monkeypatch: pytest.MonkeyPatch) -> PubSubAnalyticsPublisher:
client = _FakePublisherClient()
pubsub_v1 = types.ModuleType("google.cloud.pubsub_v1")
pubsub_v1.PublisherClient = lambda: client # type: ignore[attr-defined]
google_cloud = types.ModuleType("google.cloud")
google_cloud.pubsub_v1 = pubsub_v1 # type: ignore[attr-defined]
google = types.ModuleType("google")
google.cloud = google_cloud # type: ignore[attr-defined]
monkeypatch.setitem(sys.modules, "google", google)
monkeypatch.setitem(sys.modules, "google.cloud", google_cloud)
monkeypatch.setitem(sys.modules, "google.cloud.pubsub_v1", pubsub_v1)
monkeypatch.setenv("PUBSUB_EXCLUDED_EVENT_TYPES", "GRL.NATIVE_OUTPUT_GUARDRAILS")
monkeypatch.setenv("PUBSUB_PAYLOAD_MODE", "legacy")
publisher = PubSubAnalyticsPublisher(topic_path="projects/test/topics/analytics")
publisher.client = client
return publisher
@pytest.mark.asyncio
async def test_excluded_event_is_not_sent_to_pubsub(pubsub_publisher: PubSubAnalyticsPublisher) -> None:
await pubsub_publisher.publish("GRL.NATIVE_OUTPUT_GUARDRAILS", {"session_id": "session-1"})
assert pubsub_publisher.client.calls == []
@pytest.mark.asyncio
async def test_non_excluded_event_is_sent_to_pubsub(pubsub_publisher: PubSubAnalyticsPublisher) -> None:
await pubsub_publisher.publish("GRL.002", {"session_id": "session-1"})
assert len(pubsub_publisher.client.calls) == 1
topic_path, _, attributes = pubsub_publisher.client.calls[0]
assert topic_path == "projects/test/topics/analytics"
assert attributes["event_type"] == "GRL.002"

View File

@@ -0,0 +1,12 @@
import pytest
from types import SimpleNamespace
from agent_framework.rag.rag_service import RagService
@pytest.mark.asyncio
async def test_rag_service_retrieves_relevant_document():
settings = SimpleNamespace(VECTOR_STORE_PROVIDER='memory', GRAPH_STORE_PROVIDER='memory', RAG_TOP_K=2)
rag = RagService(settings)
await rag.add_documents(['fatura alta por roaming internacional', 'pedido de troca de aparelho'], namespace='billing')
result = await rag.retrieve('minha fatura veio alta', namespace='billing')
assert result.documents
assert 'fatura' in result.as_prompt_context().lower()

View File

@@ -0,0 +1,6 @@
from agent_framework.rag.graph_store import InMemoryGraphStore
def test_inmemory_graph_has_pgql_method_for_interface_parity():
graph = InMemoryGraphStore()
assert hasattr(graph, "pgql")

View File

@@ -0,0 +1,48 @@
import pytest
from agent_framework.checkpoints.checkpoint_repository import (
CheckpointRecoveryError,
InMemoryCheckpointRepository,
ResilientCheckpointRepository,
)
@pytest.mark.asyncio
async def test_integrity_envelope_and_recovery_skips_corrupt_latest():
raw = InMemoryCheckpointRepository()
repo = ResilientCheckpointRepository(raw, compact_every=100, keep_last=10, recovery_scan_limit=5)
await repo.put("thread-1", {"checkpoint_id": "ok-1", "checkpoint": {"id": "ok-1", "value": 1}})
await repo.put("thread-1", {"checkpoint_id": "ok-2", "checkpoint": {"id": "ok-2", "value": 2}})
# Simula corrupção no último registro persistido.
raw._data["thread-1"][-1]["payload"]["checkpoint"]["value"] = 999
recovered = await repo.get_latest("thread-1")
assert recovered["checkpoint_id"] == "ok-1"
assert recovered["checkpoint"]["value"] == 1
@pytest.mark.asyncio
async def test_compaction_keeps_last_n_checkpoints():
raw = InMemoryCheckpointRepository()
repo = ResilientCheckpointRepository(raw, compact_every=1, keep_last=3, recovery_scan_limit=10)
for i in range(7):
await repo.put("thread-compact", {"checkpoint_id": f"cp-{i}", "checkpoint": {"id": f"cp-{i}"}})
assert len(raw._data["thread-compact"]) <= 3
latest = await repo.get_latest("thread-compact")
assert latest["checkpoint_id"] == "cp-6"
@pytest.mark.asyncio
async def test_recovery_raises_when_only_corrupt_checkpoints_exist():
raw = InMemoryCheckpointRepository()
repo = ResilientCheckpointRepository(raw, compact_every=100, keep_last=10, recovery_scan_limit=5)
await repo.put("thread-bad", {"checkpoint_id": "bad", "checkpoint": {"id": "bad", "value": 1}})
raw._data["thread-bad"][-1]["payload"]["checkpoint"]["value"] = 2
with pytest.raises(CheckpointRecoveryError):
await repo.get_latest("thread-bad")

View File

@@ -0,0 +1,204 @@
from __future__ import annotations
import json
from types import SimpleNamespace
import pytest
from agent_framework.routing.continuity import SemanticRouteContinuity
from agent_framework.routing.models import IntentDefinition
class FakeLLM:
def __init__(self, response: dict | str):
self.response = response
self.calls = []
async def ainvoke(self, messages, **kwargs):
self.calls.append((messages, kwargs))
if isinstance(self.response, str):
return self.response
return json.dumps(self.response)
class FakeTelemetry:
def __init__(self):
self.events = []
async def event(self, name, payload):
self.events.append((name, payload))
def settings(**overrides):
values = {
"ENABLE_ROUTE_STICKINESS": True,
"ROUTE_STICKINESS_LLM_PROFILE": "route_continuity",
"ROUTE_STICKINESS_CONFIDENCE_THRESHOLD": 0.90,
"ROUTE_STICKINESS_HISTORY_TURNS": 2,
"ROUTE_STICKINESS_MAX_TOKENS": 80,
}
values.update(overrides)
return SimpleNamespace(**values)
def intents():
return [
IntentDefinition(
name="product_services_information",
agent="product_agent",
description="Planos, serviços, benefícios e mudança de plano.",
),
IntentDefinition(
name="billing_invoice_explanation",
agent="billing_agent",
description="Faturas, pagamentos, cobranças e contestação.",
),
]
def state(message="o que está incluso?"):
return {
"session_id": "s1",
"active_agent": "product_agent",
"intent": "product_services_information",
"domain": "telecom",
"route_decision": {
"intent": "product_services_information",
"domain": "telecom",
"mcp_tools": ["consultar_plano"],
},
"history": [
{"role": "user", "content": "qual é o meu plano?"},
{"role": "assistant", "content": "Seu plano atual é Controle 50GB."},
],
"user_text": message,
"sanitized_input": message,
}
@pytest.mark.asyncio
async def test_continue_bypasses_router_without_regex_rules():
llm = FakeLLM({"decision": "CONTINUE", "confidence": 0.97, "reason": "Continua o assunto do plano."})
telemetry = FakeTelemetry()
policy = SemanticRouteContinuity(settings(), llm, telemetry)
decision = await policy.evaluate(state(), intents=intents())
assert decision is not None
assert decision.agent == "product_agent"
assert decision.method == "continuity"
assert decision.metadata["route_bypassed"] is True
assert llm.calls[0][1]["profile_name"] == "route_continuity"
prompt = json.loads(llm.calls[0][0][1]["content"])
assert prompt["current_message"] == "o que está incluso?"
assert "product_agent" not in prompt["other_agents"]
assert telemetry.events[-1][1]["route_bypassed"] is True
@pytest.mark.asyncio
async def test_route_result_falls_back_to_enterprise_router():
llm = FakeLLM({"decision": "ROUTE", "confidence": 0.98, "reason": "Novo assunto de cobrança."})
policy = SemanticRouteContinuity(settings(), llm)
decision = await policy.evaluate(
state("agora quero contestar uma cobrança"), intents=intents()
)
assert decision is None
@pytest.mark.asyncio
async def test_low_confidence_continue_falls_back_safely():
llm = FakeLLM({"decision": "CONTINUE", "confidence": 0.70, "reason": "Possível continuidade."})
policy = SemanticRouteContinuity(settings(), llm)
assert await policy.evaluate(state(), intents=intents()) is None
@pytest.mark.asyncio
async def test_invalid_output_falls_back_safely():
llm = FakeLLM("not-json")
policy = SemanticRouteContinuity(settings(), llm)
assert await policy.evaluate(state(), intents=intents()) is None
@pytest.mark.asyncio
async def test_no_active_agent_still_classifies_global_session_actions():
llm = FakeLLM({"decision": "ROUTE", "confidence": 1.0})
policy = SemanticRouteContinuity(settings(), llm)
current = state()
current.pop("active_agent")
assert await policy.evaluate(current, intents=intents()) is None
assert len(llm.calls) == 1
@pytest.mark.asyncio
async def test_human_handoff_is_returned_as_global_route():
llm = FakeLLM({
"decision": "HUMAN_HANDOFF",
"confidence": 0.99,
"reason": "O usuário pediu atendimento humano.",
})
policy = SemanticRouteContinuity(settings(), llm)
decision = await policy.evaluate(
state("quero falar com um atendente"), intents=intents()
)
assert decision is not None
assert decision.route == "human_handoff"
assert decision.agent == "human_handoff"
assert decision.intent == "human_handoff"
assert decision.handoff is True
assert decision.metadata["session_control"] == "HUMAN_HANDOFF"
assert decision.metadata["route_bypassed"] is True
@pytest.mark.asyncio
async def test_end_session_is_returned_as_global_route():
llm = FakeLLM({
"decision": "END_SESSION",
"confidence": 0.98,
"reason": "O usuário informou que não precisa continuar.",
})
policy = SemanticRouteContinuity(settings(), llm)
decision = await policy.evaluate(state("obrigado, era só isso"), intents=intents())
assert decision is not None
assert decision.route == "end_session"
assert decision.agent == "end_session"
assert decision.intent == "end_session"
assert decision.handoff is False
assert decision.metadata["session_control"] == "END_SESSION"
assert decision.metadata["route_bypassed"] is True
@pytest.mark.asyncio
async def test_global_session_actions_work_without_active_agent():
llm = FakeLLM({
"decision": "HUMAN_HANDOFF",
"confidence": 0.97,
"reason": "Solicitação explícita de pessoa.",
})
policy = SemanticRouteContinuity(settings(), llm)
current = state("quero uma pessoa")
current.pop("active_agent")
decision = await policy.evaluate(current, intents=intents())
assert decision is not None
assert decision.route == "human_handoff"
assert len(llm.calls) == 1
@pytest.mark.asyncio
async def test_continue_without_active_agent_falls_back_to_router():
llm = FakeLLM({"decision": "CONTINUE", "confidence": 0.99})
policy = SemanticRouteContinuity(settings(), llm)
current = state()
current.pop("active_agent")
assert await policy.evaluate(current, intents=intents()) is None
assert len(llm.calls) == 1

View File

@@ -0,0 +1,19 @@
import pytest
from types import SimpleNamespace
from agent_framework.sse.events import SSEEvent, SSEHub
@pytest.mark.asyncio
async def test_sse_event_encoding():
encoded = SSEEvent(event='message', data={'text':'ok'}, id=10).encode()
assert 'id: 10' in encoded
assert 'event: message' in encoded
assert 'data: {"text": "ok"}' in encoded
@pytest.mark.asyncio
async def test_sse_hub_emit_and_replay(tmp_path):
settings = SimpleNamespace(SQLITE_DB_PATH=str(tmp_path/'db.sqlite'), SSE_KEEPALIVE_SECONDS=0.1, SSE_EVENT_REPLAY_LIMIT=10, SESSION_REPOSITORY_PROVIDER='sqlite', SSE_STORE_PROVIDER='sqlite')
hub = SSEHub(settings)
eid = await hub.emit('s1', 'flow.start', {'a': 1})
replayed = await hub.replay('s1', 0)
assert replayed[0].id == eid
assert replayed[0].event == 'flow.start'

View File

@@ -0,0 +1,30 @@
import pytest
from types import SimpleNamespace
from agent_framework.sse.events import SSEHub, SSEEvent
class MemorySSEStore:
def __init__(self):
self.rows=[]; self.next_id=1
def append_sse_event(self, session_id, event, payload):
row={"id": self.next_id, "session_id": session_id, "event_name": event, "payload": payload}
self.next_id += 1; self.rows.append(row); return row["id"]
def list_sse_events(self, session_id, after_id, limit):
return [r for r in self.rows if r["session_id"] == session_id and r["id"] > after_id][:limit]
@pytest.mark.asyncio
async def test_subscribe_skips_live_event_already_replayed():
hub = SSEHub(SimpleNamespace(SSE_KEEPALIVE_SECONDS=0.01, SSE_EVENT_REPLAY_LIMIT=100, SQLITE_DB_PATH=':memory:', SESSION_REPOSITORY_PROVIDER='sqlite'), telemetry=None)
hub.store = MemorySSEStore()
eid = await hub.emit("s1", "message.responded", {"text":"ok"})
# Same event remains in live queue and is also in replay store.
gen = hub.subscribe("s1", 0)
chunks=[]
chunks.append(await gen.__anext__()) # replay event
chunks.append(await gen.__anext__()) # connected
chunks.append(await gen.__anext__()) # keepalive, not duplicated event
assert chunks[0].startswith(f"id: {eid}")
assert "message.responded" in chunks[0]
assert "message.responded" not in chunks[2]
await gen.aclose()

View File

@@ -0,0 +1,289 @@
from __future__ import annotations
import pytest
from agent_framework.analytics.providers.langfuse import LangfuseAnalyticsPublisher
from agent_framework.observability.context import clear_observability_context, set_observability_context
from agent_framework.observability.telemetry import Telemetry
class Settings:
ENABLE_LANGFUSE = False
ENABLE_OTEL = False
LANGFUSE_TRACE_MODE = "compact"
class FakeObservation:
_next_id = 1
def __init__(self, kwargs):
self.kwargs = kwargs
self.id = f"obs-{FakeObservation._next_id}"
self.trace_id = "trace-123"
FakeObservation._next_id += 1
self.updates = []
self.trace_updates = []
self.trace_io_updates = []
def update(self, **kwargs):
self.updates.append(kwargs)
def update_trace(self, **kwargs):
self.trace_updates.append(kwargs)
def set_trace_io(self, **kwargs):
self.trace_io_updates.append(kwargs)
class FakeContextManager:
def __init__(self, observation):
self.observation = observation
def __enter__(self):
return self.observation
def __exit__(self, exc_type, exc, tb):
return False
class FakePropagationContext:
def __init__(self, owner, kwargs):
self.owner = owner
self.kwargs = kwargs
def __enter__(self):
self.owner.propagations.append(self.kwargs)
def __exit__(self, exc_type, exc, tb):
return False
class FakeLangfuse:
def __init__(self, *, legacy_api: bool = False):
self.observations = []
self.propagations = []
self.trace_updates = []
self.flush_count = 0
self.api = FakeApi() if legacy_api else None
def start_as_current_observation(self, **kwargs):
observation = FakeObservation(kwargs)
self.observations.append(observation)
return FakeContextManager(observation)
def propagate_attributes(self, **kwargs):
return FakePropagationContext(self, kwargs)
def update_current_trace(self, **kwargs):
self.trace_updates.append(kwargs)
def flush(self):
self.flush_count += 1
class FakeIngestionResponse:
errors = []
successes = []
class FakeIngestion:
def __init__(self):
self.batches = []
def batch(self, *, batch, metadata=None):
self.batches.append({"batch": batch, "metadata": metadata})
return FakeIngestionResponse()
class FakeApi:
def __init__(self):
self.ingestion = FakeIngestion()
def telemetry_with_fake_langfuse(*, legacy_api: bool = False):
FakeObservation._next_id = 1
telemetry = Telemetry(Settings())
telemetry.enabled = True
telemetry.langfuse = FakeLangfuse(legacy_api=legacy_api)
return telemetry
@pytest.mark.asyncio
async def test_compact_keeps_root_output_and_shows_ic_aga_noc_as_spans():
clear_observability_context()
telemetry = telemetry_with_fake_langfuse()
async with telemetry.span("agent.gateway_message", session_id="s1", input={"request": "cms"}, _root_span=True) as span:
await telemetry.event("IC.INTERNAL", {"step": "visible"}, kind="ic")
await telemetry.event("NOC.001", {"step": "visible"}, kind="noc")
await telemetry.event("AGA.010", {"step": "visible"}, kind="ic")
span.set_output({"answer": "ok"})
names = [obs.kwargs["name"] for obs in telemetry.langfuse.observations]
assert names == ["agent.gateway_message", "IC.INTERNAL", "NOC.001", "AGA.010"]
root = telemetry.langfuse.observations[0]
assert root.updates[-1]["input"] == {"request": "cms"}
assert root.updates[-1]["output"] == {"answer": "ok"}
assert root.trace_io_updates[-1] == {"input": {"request": "cms"}, "output": {"answer": "ok"}}
for observation in telemetry.langfuse.observations[1:]:
assert observation.kwargs.get("trace_context") is None
assert observation.kwargs["as_type"] == "span"
ic = telemetry.langfuse.observations[1]
assert ic.kwargs["input"]["step"] == "visible"
assert ic.updates[-1]["input"]["step"] == "visible"
assert ic.updates[-1]["output"] == {"status": "ok"}
assert telemetry.langfuse.propagations[-1]["trace_name"] == "agent.gateway_message"
aggregated = root.updates[-1]["metadata"]["aggregated_events"]
assert [event["name"] for event in aggregated] == ["IC.INTERNAL", "NOC.001", "AGA.010"]
@pytest.mark.asyncio
async def test_analytics_control_event_is_a_span_and_not_a_trace_tag():
clear_observability_context()
set_observability_context(request_id="req-1", trace_id="req-1", session_id="s1")
langfuse = FakeLangfuse()
publisher = LangfuseAnalyticsPublisher(langfuse=langfuse)
envelope = {
"eventType": "IC.ORDER_CONFIRMED",
"source": "agent_framework",
"payload": {"tag": "IC.ORDER_CONFIRMED", "order_id": "order-1"},
"metadata": {"ic": True},
}
await publisher.publish("IC.ORDER_CONFIRMED", envelope)
assert [obs.kwargs["name"] for obs in langfuse.observations] == ["IC.ORDER_CONFIRMED"]
observation = langfuse.observations[0]
assert observation.kwargs["as_type"] == "span"
assert observation.kwargs["input"] == envelope
assert observation.updates[-1]["output"] == {"published": True}
assert len(langfuse.trace_updates) == 1
assert "tags" not in langfuse.trace_updates[0]
@pytest.mark.asyncio
async def test_compact_generation_records_io_model_parameters_and_usage_details():
clear_observability_context()
telemetry = telemetry_with_fake_langfuse()
async with telemetry.span("agent.gateway_message", session_id="s1", input={"request": "cms"}, _root_span=True):
async with telemetry.generation_span(
name="llm.test",
model="test-model",
input=[{"role": "user", "content": "ping"}],
metadata={"profile_name": "test"},
model_parameters={"temperature": 0.2, "max_tokens": 100},
) as generation:
generation.set_output("pong")
generation.set_usage({"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2, "cost_usd": 0.01})
generation = telemetry.langfuse.observations[1]
assert generation.kwargs["name"] == "llm.test"
assert generation.kwargs["as_type"] == "generation"
assert generation.kwargs["input"] == [{"role": "user", "content": "ping"}]
assert generation.kwargs["model"] == "test-model"
assert generation.kwargs["model_parameters"] == {"temperature": 0.2, "max_tokens": 100}
assert "usage" not in generation.kwargs
assert "usage_details" not in generation.kwargs
assert generation.updates[-1]["input"] == [{"role": "user", "content": "ping"}]
assert generation.updates[-1]["output"] == "pong"
assert generation.updates[-1]["usage_details"] == {"input": 1, "output": 1}
assert generation.updates[-1]["cost_details"] == {"total": 0.01}
assert generation.kwargs.get("trace_context") is None
@pytest.mark.asyncio
async def test_legacy_io_fallback_updates_same_root_and_generation_observations():
clear_observability_context()
telemetry = telemetry_with_fake_langfuse(legacy_api=True)
async with telemetry.span("agent.gateway_message", session_id="s1", input={"request": "cms"}, _root_span=True) as root:
async with telemetry.generation_span(
name="llm.test",
model="test-model",
input=[{"role": "user", "content": "ping"}],
model_parameters={"temperature": 0.2},
) as generation:
generation.set_output("pong")
generation.set_usage({"prompt_tokens": 2, "completion_tokens": 3, "total_tokens": 5})
root.set_output({"answer": "ok"})
batches = telemetry.langfuse.api.ingestion.batches
assert [event.type for item in batches for event in item["batch"]] == ["generation-update", "span-update"]
generation_event = batches[0]["batch"][0]
assert generation_event.body.id == "obs-2"
assert generation_event.body.trace_id == "trace-123"
assert generation_event.body.input == [{"role": "user", "content": "ping"}]
assert generation_event.body.output == "pong"
assert generation_event.body.usage_details == {"input": 2, "output": 3}
root_event = batches[1]["batch"][0]
assert root_event.body.id == "obs-1"
assert root_event.body.trace_id == "trace-123"
assert root_event.body.input == {"request": "cms"}
assert root_event.body.output == {"answer": "ok"}
assert len(telemetry.langfuse.observations) == 2
@pytest.mark.asyncio
async def test_langfuse_v4_module_level_propagation_sets_native_session_context():
"""SDK v4 propagation must receive the business session id natively."""
clear_observability_context()
telemetry = telemetry_with_fake_langfuse()
calls = []
def v4_propagate_attributes(**kwargs):
calls.append(kwargs)
return FakePropagationContext(telemetry.langfuse, kwargs)
# Simulates ``from langfuse import propagate_attributes`` from SDK v4.
telemetry._langfuse_propagate_attributes = v4_propagate_attributes
async with telemetry.span(
"agent.gateway_message",
session_id="default:telecom_contas:session-123",
user_id="11999999999",
agent_id="telecom_contas",
tenant_id="default",
input={"message": "hello"},
tags=["agent:telecom_contas"],
_root_span=True,
):
await telemetry.event("IC.TEST", {"ok": True}, kind="ic")
assert len(calls) == 1
assert calls[0]["session_id"] == "default:telecom_contas:session-123"
assert calls[0]["user_id"] == "11999999999"
assert calls[0]["trace_name"] == "agent.gateway_message"
assert calls[0]["metadata"]["agent_id"] == "telecom_contas"
assert calls[0]["metadata"]["tenant_id"] == "default"
assert calls[0]["tags"] == ["agent:telecom_contas"]
# Keep the legacy update as a compatibility fallback, but the v4 path above
# is now the authoritative way to materialize native sessionId.
root = telemetry.langfuse.observations[0]
assert any(
update.get("session_id") == "default:telecom_contas:session-123"
for update in root.trace_updates
)
def test_trace_attribute_propagation_keeps_legacy_client_method_fallback():
telemetry = telemetry_with_fake_langfuse()
telemetry._langfuse_propagate_attributes = None
cm = telemetry._start_trace_attribute_propagation(
"agent.gateway_message",
{
"session_id": "legacy-session",
"user_id": "legacy-user",
"agent_id": "legacy-agent",
},
)
assert cm is not None
with cm:
pass
assert telemetry.langfuse.propagations[-1]["session_id"] == "legacy-session"

View File

@@ -0,0 +1,23 @@
from types import SimpleNamespace
from agent_framework.observability.token_cost import TokenUsageCollector, TokenUsage
def test_token_usage_extracts_cached_and_reasoning_tokens():
usage = TokenUsage.from_openai_usage({
"prompt_tokens": 1000,
"completion_tokens": 500,
"total_tokens": 1600,
"prompt_tokens_details": {"cached_tokens": 250},
"completion_tokens_details": {"reasoning_tokens": 100},
})
assert usage.prompt_tokens == 1000
assert usage.cached_tokens == 250
assert usage.reasoning_tokens == 100
assert usage.total_tokens == 1600
def test_cost_tracker_uses_model_prices_json():
settings = SimpleNamespace(MODEL_PRICES_JSON='{"my-model":{"input_per_1m":"1","output_per_1m":"2","cached_input_per_1m":"0.1"}}', USD_BRL_RATE='5')
enriched = TokenUsageCollector(settings).enrich("my-model", {"prompt_tokens": 1000, "completion_tokens": 1000, "prompt_tokens_details": {"cached_tokens": 500}})
assert enriched["cost_usd"] > 0
assert abs(enriched["cost_brl"] - enriched["cost_usd"] * 5) < 1e-9

View File

@@ -0,0 +1,98 @@
from __future__ import annotations
from types import SimpleNamespace
from agent_framework.mcp.tool_policy import ToolPolicyRegistry
from agent_framework.mcp.tool_router import MCPToolRouter
class _Registry:
def __init__(self, tool=None):
self.tool = tool
def get_tool(self, _name):
return self.tool
def _router(policy_registry, legacy=None):
router = MCPToolRouter.__new__(MCPToolRouter)
router.tool_policies = policy_registry
router.registry = _Registry(legacy)
return router
def test_missing_policy_file_preserves_legacy_behavior(tmp_path):
policies = ToolPolicyRegistry(str(tmp_path / "missing.yaml"))
legacy = SimpleNamespace(
tool_type="action",
requires=["order_id"],
confirmation_required=True,
execution_policy={},
)
router = _router(policies, legacy)
allowed, reason, metadata = router.validate_execution_policy("alterar", {"order_id": "42"})
assert allowed is False
assert "confirmação" in reason
assert metadata["operation_type"] == "transactional"
assert metadata["policy_source"] == "tools.yaml"
def test_read_only_policy_executes_without_confirmation(tmp_path):
path = tmp_path / "tool_policies.yaml"
path.write_text(
"tool_policies:\n consultar:\n operation_type: read_only\n",
encoding="utf-8",
)
router = _router(ToolPolicyRegistry(str(path)))
allowed, reason, metadata = router.validate_execution_policy("consultar", {})
assert allowed is True
assert reason is None
assert metadata["operation_type"] == "read_only"
def test_transactional_policy_requires_literal_boolean_confirmation(tmp_path):
path = tmp_path / "tool_policies.yaml"
path.write_text(
"tool_policies:\n cancelar:\n operation_type: transactional\n require_confirmation: true\n",
encoding="utf-8",
)
router = _router(ToolPolicyRegistry(str(path)))
denied, _, _ = router.validate_execution_policy("cancelar", {"confirmed": "true"})
allowed, reason, metadata = router.validate_execution_policy("cancelar", {"confirmed": True})
assert denied is False
assert allowed is True
assert reason is None
assert metadata["policy_source"] == "tool_policies.yaml"
def test_requires_confirmation_alias_is_supported(tmp_path):
path = tmp_path / "tool_policies.yaml"
path.write_text(
"tool_policies:\n alterar:\n type: transactional\n requires_confirmation: true\n",
encoding="utf-8",
)
policy = ToolPolicyRegistry(str(path)).get("alterar")
assert policy.operation_type == "transactional"
assert policy.require_confirmation is True
def test_workflow_execution_policy_is_exposed(tmp_path):
path = tmp_path / "tool_policies.yaml"
path.write_text(
"""defaults:\n operation_type: read_only\ntool_policies:\n refund:\n operation_type: transactional\n require_confirmation: true\n execution:\n mode: workflow\n workflow: refund_order\n version: 2\n""",
encoding="utf-8",
)
registry = ToolPolicyRegistry(str(path))
policy = registry.get("refund")
assert policy is not None
assert policy.execution.mode == "workflow"
assert policy.execution.workflow == "refund_order"
assert policy.execution.version == 2

View File

@@ -0,0 +1,110 @@
from pathlib import Path
import pytest
from agent_framework.workflows import (
FileWorkflowRepository,
WorkflowActionRegistry,
WorkflowRuntime,
WorkflowToolExecutor,
)
@pytest.mark.asyncio
async def test_deterministic_workflow_routes_and_caches(tmp_path: Path):
(tmp_path / "refund.active.yaml").write_text("version: 1\n", encoding="utf-8")
(tmp_path / "refund.v1.yaml").write_text(
"""name: refund
version: 1
start: validate
nodes:
- id: validate
action: validate
input: {order_id: $.input.order_id}
- id: execute
action: execute
input: {order_id: $.input.order_id}
edges:
- from: validate
to: execute
when: {path: $.nodes.validate.valid, equals: true}
- from: validate
to: END
when: {path: $.nodes.validate.valid, equals: false}
- from: execute
to: END
""",
encoding="utf-8",
)
actions = WorkflowActionRegistry()
actions.register("validate", lambda params, state: {"valid": params["order_id"] == "123"})
actions.register("execute", lambda params, state: {"protocol": "P-1"})
runtime = WorkflowRuntime(FileWorkflowRepository(tmp_path), actions=actions)
ok = await runtime.arun("refund", {"order_id": "123"})
assert ok.status == "COMPLETED"
assert ok.output["execute"]["protocol"] == "P-1"
assert len(runtime._compiled) == 1
rejected = await runtime.arun("refund", {"order_id": "999"})
assert rejected.status == "COMPLETED"
assert "execute" not in rejected.output
assert len(runtime._compiled) == 1
@pytest.mark.asyncio
async def test_policy_adapter_runs_only_workflow_mode(tmp_path: Path):
(tmp_path / "job.active.yaml").write_text("version: 1\n", encoding="utf-8")
(tmp_path / "job.v1.yaml").write_text(
"""name: job
version: 1
start: one
nodes:
- id: one
action: one
edges:
- from: one
to: END
""",
encoding="utf-8",
)
actions = WorkflowActionRegistry()
actions.register("one", lambda params, state: {"ok": True})
adapter = WorkflowToolExecutor(WorkflowRuntime(FileWorkflowRepository(tmp_path), actions=actions))
assert await adapter.execute_from_policy(tool_name="x", arguments={}, policy={"execution": {"mode": "direct_tool"}}) is None
result = await adapter.execute_from_policy(tool_name="x", arguments={}, policy={"execution": {"mode": "workflow", "workflow": "job", "version": "active"}})
assert result["status"] == "COMPLETED"
@pytest.mark.asyncio
async def test_offline_regression_mode_forces_deterministic_backend_even_if_langgraph_is_available(tmp_path: Path, monkeypatch):
(tmp_path / "offline.active.yaml").write_text("version: 1\n", encoding="utf-8")
(tmp_path / "offline.v1.yaml").write_text(
"""name: offline
version: 1
start: one
nodes:
- id: one
action: one
edges:
- from: one
to: END
""",
encoding="utf-8",
)
actions = WorkflowActionRegistry()
actions.register("one", lambda params, state: {"ok": True})
runtime = WorkflowRuntime(
FileWorkflowRepository(tmp_path),
actions=actions,
allow_deterministic_fallback=True,
)
def _must_not_compile(_definition):
raise AssertionError("offline regression mode must not compile LangGraph")
monkeypatch.setattr(runtime, "_compile", _must_not_compile)
result = await runtime.arun("offline", {})
assert result.status == "COMPLETED"
assert result.output["one"]["ok"] is True
assert runtime._compiled == {}

View File

@@ -0,0 +1,14 @@
from pathlib import Path
ROOT = Path(__file__).resolve().parents[2]
WORKFLOW = ROOT / 'agent_template_backend' / 'app' / 'workflows' / 'agent_graph.py'
def test_workflow_uses_framework_checkpointer_not_memory_saver():
src = WORKFLOW.read_text()
assert 'create_langgraph_checkpointer(self.settings)' in src
assert 'MemorySaver()' not in src
def test_workflow_wraps_nodes_with_langgraph_telemetry():
src = WORKFLOW.read_text()
assert 'self._node("input_guardrails", self.input_guardrails)' in src
assert 'async with self.langgraph_telemetry.node(name, state)' in src