From a3f3f047dac871acb91be1701a78efc603b5cc8c Mon Sep 17 00:00:00 2001
From: Jon Saad-Falcon <41205309+jonsaadfalcon@users.noreply.github.com>
Date: Mon, 16 Mar 2026 20:41:24 -0700
Subject: [PATCH] feat: add FakeEngine and scenario_harness fixture for agent
 lifecycle testing

---
 tests/agents/conftest.py                    | 47 ++++++++++++++
 tests/agents/fake_engine.py                 | 70 +++++++++++++++++++++
 tests/agents/scenario_harness.py            | 38 +++++++++++
 tests/agents/test_fake_engine.py            | 42 +++++++++++++
 tests/agents/test_scenario_harness_smoke.py | 19 ++++++
 5 files changed, 216 insertions(+)
 create mode 100644 tests/agents/conftest.py
 create mode 100644 tests/agents/fake_engine.py
 create mode 100644 tests/agents/scenario_harness.py
 create mode 100644 tests/agents/test_fake_engine.py
 create mode 100644 tests/agents/test_scenario_harness_smoke.py

diff --git a/tests/agents/conftest.py b/tests/agents/conftest.py
new file mode 100644
index 00000000..41277334
--- /dev/null
+++ b/tests/agents/conftest.py
@@ -0,0 +1,47 @@
+"""Shared fixtures for agent tests."""
+
+from __future__ import annotations
+
+import pytest
+
+from openjarvis.agents.executor import AgentExecutor
+from openjarvis.agents.manager import AgentManager
+from openjarvis.agents.scheduler import AgentScheduler
+from openjarvis.core.events import EventBus
+from tests.agents.fake_engine import FakeEngine
+from tests.agents.scenario_harness import FakeSystem, ScenarioHarness
+
+
+@pytest.fixture
+def scenario_harness(tmp_path):
+    """Wire up real components for agent lifecycle testing."""
+    from openjarvis.agents.monitor_operative import MonitorOperativeAgent
+    from openjarvis.core.registry import AgentRegistry
+
+    # Re-register agent types (conftest auto-clears registries)
+    if not AgentRegistry.contains("monitor_operative"):
+        AgentRegistry.register("monitor_operative")(MonitorOperativeAgent)
+
+    db_path = str(tmp_path / "agents.db")
+    bus = EventBus(record_history=True)
+    manager = AgentManager(db_path=db_path)
+
+    engine = FakeEngine([{"content": "Default response."}])
+    system = FakeSystem(engine=engine)
+
+    executor = AgentExecutor(manager=manager, event_bus=bus)
+    executor.set_system(system)
+
+    scheduler = AgentScheduler(
+        manager=manager, executor=executor, event_bus=bus,
+    )
+
+    return ScenarioHarness(
+        manager=manager,
+        executor=executor,
+        scheduler=scheduler,
+        bus=bus,
+        engine=engine,
+        system=system,
+        db_path=db_path,
+    )
diff --git a/tests/agents/fake_engine.py b/tests/agents/fake_engine.py
new file mode 100644
index 00000000..805b05a4
--- /dev/null
+++ b/tests/agents/fake_engine.py
@@ -0,0 +1,70 @@
+"""Scripted inference engine for deterministic agent testing."""
+
+from __future__ import annotations
+
+from typing import Any, AsyncIterator, Dict, List
+
+from openjarvis.engine._stubs import InferenceEngine
+
+
+class FakeEngine(InferenceEngine):
+    """Returns pre-defined responses in order. Captures prompts for assertions."""
+
+    engine_id = "fake"
+
+    def __init__(self, responses: list[dict]) -> None:
+        self._responses = list(responses)
+        self._call_count = 0
+        self._last_messages: list | None = None
+
+    @property
+    def call_count(self) -> int:
+        return self._call_count
+
+    @property
+    def last_messages(self) -> list | None:
+        return self._last_messages
+
+    def generate(
+        self,
+        messages: list,
+        *,
+        model: str,
+        temperature: float = 0.7,
+        max_tokens: int = 1024,
+        **kw: Any,
+    ) -> Dict[str, Any]:
+        self._last_messages = messages
+        idx = min(self._call_count, len(self._responses) - 1)
+        resp = self._responses[idx]
+        self._call_count += 1
+
+        # Support raising exceptions for error testing
+        if "raise" in resp:
+            raise resp["raise"]
+
+        result: Dict[str, Any] = {
+            "content": resp.get("content", ""),
+            "usage": resp.get("usage", {
+                "prompt_tokens": 10,
+                "completion_tokens": 5,
+                "total_tokens": 15,
+            }),
+            "model": model,
+            "finish_reason": "tool_calls" if resp.get("tool_calls") else "stop",
+        }
+        if resp.get("tool_calls"):
+            result["tool_calls"] = resp["tool_calls"]
+        return result
+
+    async def stream(
+        self, messages: list, *, model: str, **kw: Any,
+    ) -> AsyncIterator[str]:
+        result = self.generate(messages, model=model, **kw)
+        yield result["content"]
+
+    def list_models(self) -> List[str]:
+        return ["fake-model"]
+
+    def health(self) -> bool:
+        return True
diff --git a/tests/agents/scenario_harness.py b/tests/agents/scenario_harness.py
new file mode 100644
index 00000000..0e387175
--- /dev/null
+++ b/tests/agents/scenario_harness.py
@@ -0,0 +1,38 @@
+"""Shared harness dataclasses for agent lifecycle scenario tests."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+from openjarvis.agents.executor import AgentExecutor
+from openjarvis.agents.manager import AgentManager
+from openjarvis.agents.scheduler import AgentScheduler
+from openjarvis.core.events import EventBus
+from tests.agents.fake_engine import FakeEngine
+
+
+@dataclass(slots=True)
+class FakeSystem:
+    """Lightweight stand-in for JarvisSystem — just engine + model."""
+
+    engine: FakeEngine
+    model: str = "fake-model"
+    memory_backend: Any = None
+    channel_backend: Any = None
+    tools: list = field(default_factory=list)
+    config: Any = None
+    session_store: Any = None
+
+
+@dataclass(slots=True)
+class ScenarioHarness:
+    """All components needed for an agent lifecycle scenario test."""
+
+    manager: AgentManager
+    executor: AgentExecutor
+    scheduler: AgentScheduler
+    bus: EventBus
+    engine: FakeEngine
+    system: FakeSystem
+    db_path: str
diff --git a/tests/agents/test_fake_engine.py b/tests/agents/test_fake_engine.py
new file mode 100644
index 00000000..e0815612
--- /dev/null
+++ b/tests/agents/test_fake_engine.py
@@ -0,0 +1,42 @@
+"""Tests for FakeEngine."""
+
+from __future__ import annotations
+
+import pytest
+
+from tests.agents.fake_engine import FakeEngine
+
+
+def test_fake_engine_returns_responses_in_order():
+    engine = FakeEngine([
+        {"content": "first"},
+        {"content": "second"},
+    ])
+    r1 = engine.generate([], model="m")
+    r2 = engine.generate([], model="m")
+    assert r1["content"] == "first"
+    assert r2["content"] == "second"
+    assert engine.call_count == 2
+
+
+def test_fake_engine_repeats_last_response():
+    engine = FakeEngine([{"content": "only"}])
+    engine.generate([], model="m")
+    r = engine.generate([], model="m")
+    assert r["content"] == "only"
+
+
+def test_fake_engine_raises_on_request():
+    engine = FakeEngine([{"raise": ValueError("boom")}])
+    with pytest.raises(ValueError, match="boom"):
+        engine.generate([], model="m")
+
+
+def test_fake_engine_tool_calls():
+    engine = FakeEngine([{
+        "content": "",
+        "tool_calls": [{"id": "1", "function": {"name": "think", "arguments": "{}"}}],
+    }])
+    r = engine.generate([], model="m")
+    assert r["finish_reason"] == "tool_calls"
+    assert len(r["tool_calls"]) == 1
diff --git a/tests/agents/test_scenario_harness_smoke.py b/tests/agents/test_scenario_harness_smoke.py
new file mode 100644
index 00000000..32776740
--- /dev/null
+++ b/tests/agents/test_scenario_harness_smoke.py
@@ -0,0 +1,19 @@
+"""Smoke test for the scenario harness fixture."""
+
+from __future__ import annotations
+
+from tests.agents.scenario_harness import ScenarioHarness
+
+
+def test_harness_creates_and_runs_agent(scenario_harness: ScenarioHarness):
+    """Verify the harness wires up correctly — create agent, run tick."""
+    h = scenario_harness
+    agent = h.manager.create_agent("Smoke Test", config={
+        "schedule_type": "manual",
+        "instruction": "Say hello.",
+    })
+    h.executor.execute_tick(agent["id"])
+    updated = h.manager.get_agent(agent["id"])
+    assert updated["status"] == "idle"
+    assert updated["total_runs"] == 1
+    assert updated["summary_memory"] != ""