feat(runtime): capture step evidence in console

tests/test_agents_collab_runner.py

@@ -5,7 +5,12 @@ from unittest.mock import MagicMock, patch
 
 from agents.collab_runner import CollaborativeTaskRunner, CollaborativeTaskRunnerConfig
 from agents.config import CollaborationConfig
-from agents.models import Observation, ReflectionAction, ReflectionOutcome, VerificationVerdict
+from agents.models import (
+    Observation,
+    ReflectionAction,
+    ReflectionOutcome,
+    VerificationVerdict,
+)
 from core.models import Bounds, Scene, SceneElement, Task
 from runtime.executor import StepResult
 from runtime.planner import PlannedStep
@@ -13,6 +18,7 @@ from runtime.task import TaskRunner, TaskRunnerConfig
 from storage.artifact_store import ArtifactStore
 from storage.task_metadata import TaskMetadataStore
 from storage.timeline import Timeline
+from tests.fakes import PNG_10X20
 from world.config import WorldConfig
 
 
@@ -20,7 +26,11 @@ def _scene() -> Scene:
     return Scene(
         width=1080,
         height=1920,
-        elements=[SceneElement(id="btn1", type="button", bounds=Bounds(10, 20, 100, 50), text="OK")],
+        elements=[
+            SceneElement(
+                id="btn1", type="button", bounds=Bounds(10, 20, 100, 50), text="OK"
+            )
+        ],
     )
 
 
@@ -55,7 +65,9 @@ def _replan_outcome() -> ReflectionOutcome:
 def _recovery_outcome() -> ReflectionOutcome:
     return ReflectionOutcome(
         replan=False,
-        action=ReflectionAction(action="swipe", description="Scroll", args={"direction": "up"}),
+        action=ReflectionAction(
+            action="swipe", description="Scroll", args={"direction": "up"}
+        ),
         reasoning="Try scrolling.",
     )
 
@@ -204,6 +216,7 @@ def test_collaborative_run_shares_bookkeeping_with_task_runner(tmp_path) -> None
         on_task_succeeded=on_task_succeeded,
         world_config=WorldConfig(enabled=True),
         config=TaskRunnerConfig(max_steps=5),
+        screenshot_provider=lambda device_id: PNG_10X20,
     )
 
     runner = CollaborativeTaskRunner(
@@ -225,6 +238,9 @@ def test_collaborative_run_shares_bookkeeping_with_task_runner(tmp_path) -> None
 
     assert result.status == "completed"
     assert len(timeline.read(task.id)) == 1
+    record = timeline.read(task.id)[0]
+    assert record["before_screenshot_path"]
+    assert record["after_screenshot_path"]
     assert metadata.get_task(task.id)["status"] == "completed"
     on_task_succeeded.assert_called_once_with(task.id, task.goal, timeline)