maskanx_cm_backend/tests/conftest.py

# -*- coding: utf-8 -*-
"""Shared fixtures for AOM tests."""

import pytest


@pytest.fixture(autouse=True)
def _no_ambient_operator_tokens(monkeypatch):
    """Never let a developer's real envs.json reach a test process.

    `adclaw/__init__.py` calls `load_envs_into_environ()` at import time, so
    whatever is saved in Settings > Environments on this machine becomes
    ambient environment for every test in the suite — not just the ones
    that opted in. Most campaign tests deliberately exercise the
    *unauthenticated* path (MASKANX_OPERATOR_TOKENS unset), so once a real
    operator token is configured for actual use, require_operator starts
    demanding a header those tests never send, and they fail with 401
    instead of the status the test is actually about. 28 tests broke this
    way the first time a real token was set.

    A test that wants a token still gets one: its own monkeypatch.setenv
    runs after this fixture in the same test and overrides it.
    """
    monkeypatch.delenv("MASKANX_OPERATOR_TOKENS", raising=False)

from adclaw.memory_agent.embeddings import FakeEmbeddingPipeline
from adclaw.memory_agent.models import AOMConfig
from adclaw.memory_agent.store import MemoryStore
from adclaw.config.config import PersonaConfig, Config, AgentsConfig


@pytest.fixture
async def aom_store():
    """In-memory SQLite store — zero I/O."""
    store = MemoryStore(":memory:", dimensions=32)
    await store.initialize()
    yield store
    await store.close()


@pytest.fixture
def fake_embedder():
    """Deterministic embedding pipeline for tests."""
    return FakeEmbeddingPipeline(dimensions=32)


@pytest.fixture
def fake_llm_caller():
    """Canned LLM: extraction → JSON, consolidation → insight text."""

    async def caller(prompt: str) -> str:
        if "extract" in prompt.lower():
            return '{"entities": ["test_entity"], "topics": ["test_topic"], "importance": 0.7}'
        if "consolidat" in prompt.lower() or "synthesiz" in prompt.lower() or "cluster" in prompt.lower():
            return "INSIGHT: Test insight about the data.\nIMPORTANCE: 0.8"
        if "memory" in prompt.lower() or "question" in prompt.lower():
            return "Based on the memories, the answer is test. [Memory #abc123]"
        return "Test response"

    return caller


@pytest.fixture
def aom_config():
    """Default AOM configuration for tests."""
    return AOMConfig(
        enabled=True,
        embedding_backend="local",
        embedding_dimensions=32,
        importance_threshold=0.3,
    )


@pytest.fixture
def sample_personas():
    """Three personas: coordinator + researcher + writer."""
    return [
        PersonaConfig(id="coordinator", name="Coordinator", is_coordinator=True, soul_md="## Role\nOrchestrate the team."),
        PersonaConfig(id="researcher", name="Mike", soul_md="## Role\nResearch and analyze."),
        PersonaConfig(id="content-writer", name="Mira", soul_md="## Role\nWrite content."),
    ]

@pytest.fixture
def persona_manager(sample_personas, tmp_path):
    """PersonaManager with 3 personas and temp working dir."""
    from adclaw.agents.persona_manager import PersonaManager
    mgr = PersonaManager(working_dir=str(tmp_path), personas=sample_personas)
    mgr.ensure_dirs()
    return mgr

@pytest.fixture
def config_with_personas(sample_personas):
    """Config object with 3 personas."""
    return Config(agents=AgentsConfig(personas=sample_personas))