CubeLoop — a Pythonic, async-native agent harness core designed for high performance, readability, and production-grade persistence.
Python agent-frameworkagent-harnessagent-runtimeagentic-aiai-agentsai-frameworkanthropicharnessharness-engineeringharness-frameworklanggraphlanggraph-alternativellmmcpopenaipi-agentpi-agent-corepythonpython-agent
import pytest
from cubeloop.providers.images.base import BaseImagesProvider, ImagesProvider
from cubeloop.providers.images.capability import (
ImagesCapabilityDescriptor,
SizeSpec,
)
from cubeloop.providers.images.types import (
AssistantImages,
ImagesContext,
ImagesCost,
ImagesModel,
)
class _StubBase(BaseImagesProvider):
"""Minimal concrete subclass exposing helpers + a no-op generate_images."""
async def generate_images(self, model, context, *, options=None):
return AssistantImages(
api=model.api,
provider_id=model.provider_id,
model=model.id,
output=[],
)
def test_provider_id_stored_on_instance():
p = _StubBase(provider_id="openai")
assert p.provider_id == "openai"
def test_default_capability_is_openai_shape():
p = _StubBase(provider_id="openai")
# _capability_for falls back to the default descriptor when no override matches.
cap = p._capability_for(ImagesModel(id="gpt-image-1", provider_id="openai"))
assert cap.size_spec.kind == "size_string"
def test_capability_override_by_model_id():
base_cap = ImagesCapabilityDescriptor()
flux_cap = ImagesCapabilityDescriptor(size_spec=SizeSpec(kind="aspect_ratio"))
p = _StubBase(
provider_id="together",
capability=base_cap,
model_capability_overrides={"flux-schnell": flux_cap},
)
assert (
p._capability_for(
ImagesModel(id="flux-schnell", provider_id="together")
).size_spec.kind
== "aspect_ratio"
)
assert (
p._capability_for(
ImagesModel(id="flux-pro", provider_id="together")
).size_spec.kind
== "size_string"
)
def test_model_factory_propagates_provider_id():
p = _StubBase(provider_id="doubao")
model = p.model("doubao-seedream-4-5-251128", api="doubao-images")
assert isinstance(model, ImagesModel)
assert model.id == "doubao-seedream-4-5-251128"
assert model.provider_id == "doubao"
assert model.api == "doubao-images"
assert model.default_size is None # nothing passed → None
def test_model_factory_passes_defaults_through():
p = _StubBase(provider_id="openai")
model = p.model(
"gpt-image-1",
default_size="1024x1024",
default_n=2,
default_quality="high",
default_output_format="png",
cost=ImagesCost(per_image=0.04),
max_input_images=4,
)
assert model.default_size == "1024x1024"
assert model.default_n == 2
assert model.default_quality == "high"
assert model.default_output_format == "png"
assert model.cost is not None
assert model.cost.per_image == 0.04
assert model.max_input_images == 4
def test_subscribe_request_and_detach():
p = _StubBase(provider_id="openai")
events: list[dict] = []
def cb(payload, model):
events.append(payload)
detach = p.subscribe_request(cb)
assert cb in p._request_listeners
detach()
assert cb not in p._request_listeners
def test_subscribe_response_and_detach():
p = _StubBase(provider_id="openai")
events: list[BaseException | None] = []
def cb(body, model, exc):
events.append(exc)
detach = p.subscribe_response(cb)
assert cb in p._response_listeners
detach()
assert cb not in p._response_listeners
def test_no_subscribe_chunk_method():
# Image is one-shot: there are no chunks, so no subscribe_chunk.
p = _StubBase(provider_id="openai")
assert not hasattr(p, "subscribe_chunk")
assert not hasattr(p, "_chunk_listeners")
def test_base_generate_images_raises_not_implemented():
base = BaseImagesProvider(provider_id="x")
with pytest.raises(NotImplementedError):
# Run as coroutine; we expect NotImplementedError at call time.
import asyncio
asyncio.run(
base.generate_images(
ImagesModel(id="x", provider_id="x"),
ImagesContext(prompt="x"),
)
)
def test_protocol_runtime_check_accepts_stub():
p = _StubBase(provider_id="openai")
assert isinstance(p, ImagesProvider)