devagent-physical-engine 0.11.4__tar.gz → 0.11.5__tar.gz

{devagent_physical_engine-0.11.4/src/devagent_physical_engine.egg-info → devagent_physical_engine-0.11.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devagent-physical-engine
-Version: 0.11.4
+Version: 0.11.5
 Summary: Verification-first agentic planning runtime for robotic and industrial automation.
 Author: Tom Ha
 License: All Rights Reserved
@@ -49,11 +49,11 @@ DevAgent Smart Physical Engine adds AI-assisted engineering to existing automati
 
 ## Current version
 
-**v0.11.4 — Beta software / evidence-gated physical runtime**
+**v0.11.5 — Beta software / evidence-gated physical runtime**
 
 The Python software stack is regression-tested on Python 3.11, 3.12, and 3.13 with separate static/coverage, package-install, and runtime dependency security gates. The UR5e Gazebo/MoveIt path now contains production-candidate continuous-clearance, tool-collision, and workpiece-lifecycle capabilities, but those physical-simulation scopes remain **EXPERIMENTAL until the executable target-workstation v0.11 campaign passes**.
 
-v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 adds live terminal progress tracing for AI request, engineering, and provider-qualification workflows. Progress is written to `stderr`; final machine-readable JSON remains isolated on `stdout`.
+v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 added live terminal progress tracing for AI request, engineering, and provider-qualification workflows. v0.11.5 adds explicit visualization modes: `--visualize` requests exact task-specific visualization and fails closed until an exact physical Twin plus compiled `PhysicalMotionPlan` exist; `--visual-demo` launches the UR5e Gazebo/MoveIt motion demo without misrepresenting that generic motion as the requested BOX/source/destination task.
 
 Real robot execution remains locked. Functional-safety certification and site qualification are not claimed.
 
@@ -66,6 +66,8 @@ Real robot execution remains locked. Functional-safety certification and site qu
 | Planner / Critic / Recovery | Implemented |
 | Natural-language engineering request front door | Implemented |
 | Live terminal stage tracing (`stderr`) | Implemented; `--quiet` supported |
+| Explicit AI visual demo (`--visual-demo`) | Implemented for UR5e workstation visual motion; not task-specific |
+| Exact task visualization (`--visualize`) | Evidence-gated; blocked until exact Twin + compiled physical motion are available |
 | Deterministic compiler / semantic / verification gates | Implemented |
 | Multi-vendor robot profile registry | Implemented |
 | UR5e / FANUC CRX / KUKA KR / ABB IRB abstraction | Implemented; physical qualification is vendor-specific |
@@ -211,7 +213,7 @@ devagent-physical-ai engineer \
   --provider openai --model <model-id>
 ```
 
-Live commands now show stage-by-stage progress while they are running, for example:
+Live commands show stage-by-stage progress while they are running, for example:
 
 ```text
 [DevAgent +   0.00s] START        provider=openai model=<model-id>
@@ -230,6 +232,30 @@ Live commands now show stage-by-stage progress while they are running, for examp
 
 Progress logs are emitted to `stderr`, so piping `stdout` still produces clean JSON. For CI or scripts that do not want terminal progress, add `--quiet`.
 
+### Visualization semantics
+
+`--visualize` means **visualize the exact engineering task**, not merely make a robot move. Therefore it fails closed until the request has enough physical evidence to build the exact scene and a compiled physical trajectory:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visualize
+```
+
+Until the task-specific physical bridge is complete, the result reports `task_visualization_requires_physical_twin_motion_plan` and lists required evidence such as robot/source/destination poses, collision geometry, workpiece/tool geometry, and a compiled `PhysicalMotionPlan`. DevAgent intentionally does not substitute a canned motion and claim it is the requested BOX task.
+
+To verify that the workstation can actually launch Gazebo/MoveIt and visibly move a UR5e now, use the explicit demo mode:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visual-demo
+```
+
+The visual demo uses the packaged UR simulation motion smoke, records an acceptance report under `~/.devagent/ai-visual`, and leaves Gazebo/MoveIt open after success unless `--close-after` is supplied. Its JSON explicitly reports `task_specific=false` and `motion_source=ur_robot_driver/example_move.py`.
+
 Equivalent routing supports Anthropic and Gemini. Live AI qualification does not unlock real robot execution.
 
 ## Software production gates

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5}/README.md

@@ -8,11 +8,11 @@ DevAgent Smart Physical Engine adds AI-assisted engineering to existing automati
 
 ## Current version
 
-**v0.11.4 — Beta software / evidence-gated physical runtime**
+**v0.11.5 — Beta software / evidence-gated physical runtime**
 
 The Python software stack is regression-tested on Python 3.11, 3.12, and 3.13 with separate static/coverage, package-install, and runtime dependency security gates. The UR5e Gazebo/MoveIt path now contains production-candidate continuous-clearance, tool-collision, and workpiece-lifecycle capabilities, but those physical-simulation scopes remain **EXPERIMENTAL until the executable target-workstation v0.11 campaign passes**.
 
-v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 adds live terminal progress tracing for AI request, engineering, and provider-qualification workflows. Progress is written to `stderr`; final machine-readable JSON remains isolated on `stdout`.
+v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 added live terminal progress tracing for AI request, engineering, and provider-qualification workflows. v0.11.5 adds explicit visualization modes: `--visualize` requests exact task-specific visualization and fails closed until an exact physical Twin plus compiled `PhysicalMotionPlan` exist; `--visual-demo` launches the UR5e Gazebo/MoveIt motion demo without misrepresenting that generic motion as the requested BOX/source/destination task.
 
 Real robot execution remains locked. Functional-safety certification and site qualification are not claimed.
 
@@ -25,6 +25,8 @@ Real robot execution remains locked. Functional-safety certification and site qu
 | Planner / Critic / Recovery | Implemented |
 | Natural-language engineering request front door | Implemented |
 | Live terminal stage tracing (`stderr`) | Implemented; `--quiet` supported |
+| Explicit AI visual demo (`--visual-demo`) | Implemented for UR5e workstation visual motion; not task-specific |
+| Exact task visualization (`--visualize`) | Evidence-gated; blocked until exact Twin + compiled physical motion are available |
 | Deterministic compiler / semantic / verification gates | Implemented |
 | Multi-vendor robot profile registry | Implemented |
 | UR5e / FANUC CRX / KUKA KR / ABB IRB abstraction | Implemented; physical qualification is vendor-specific |
@@ -170,7 +172,7 @@ devagent-physical-ai engineer \
   --provider openai --model <model-id>
 ```
 
-Live commands now show stage-by-stage progress while they are running, for example:
+Live commands show stage-by-stage progress while they are running, for example:
 
 ```text
 [DevAgent +   0.00s] START        provider=openai model=<model-id>
@@ -189,6 +191,30 @@ Live commands now show stage-by-stage progress while they are running, for examp
 
 Progress logs are emitted to `stderr`, so piping `stdout` still produces clean JSON. For CI or scripts that do not want terminal progress, add `--quiet`.
 
+### Visualization semantics
+
+`--visualize` means **visualize the exact engineering task**, not merely make a robot move. Therefore it fails closed until the request has enough physical evidence to build the exact scene and a compiled physical trajectory:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visualize
+```
+
+Until the task-specific physical bridge is complete, the result reports `task_visualization_requires_physical_twin_motion_plan` and lists required evidence such as robot/source/destination poses, collision geometry, workpiece/tool geometry, and a compiled `PhysicalMotionPlan`. DevAgent intentionally does not substitute a canned motion and claim it is the requested BOX task.
+
+To verify that the workstation can actually launch Gazebo/MoveIt and visibly move a UR5e now, use the explicit demo mode:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visual-demo
+```
+
+The visual demo uses the packaged UR simulation motion smoke, records an acceptance report under `~/.devagent/ai-visual`, and leaves Gazebo/MoveIt open after success unless `--close-after` is supplied. Its JSON explicitly reports `task_specific=false` and `motion_source=ur_robot_driver/example_move.py`.
+
 Equivalent routing supports Anthropic and Gemini. Live AI qualification does not unlock real robot execution.
 
 ## Software production gates

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "devagent-physical-engine"
-version = "0.11.4"
+version = "0.11.5"
 description = "Verification-first agentic planning runtime for robotic and industrial automation."
 readme = "README.md"
 requires-python = ">=3.11"

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5}/src/devagent_physical_engine/__init__.py

@@ -41,4 +41,4 @@ __all__ = [
     "TwinValidator",
 ]
 
-__version__ = "0.11.4"
+__version__ = "0.11.5"

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5}/src/devagent_physical_engine/ai_cli.py

@@ -4,6 +4,7 @@ import argparse
 import importlib.util
 import json
 import os
+from pathlib import Path
 import sys
 import time
 from collections.abc import Callable
@@ -45,6 +46,36 @@ def _add_live_request_args(parser: argparse.ArgumentParser) -> None:
     )
 
 
+def _add_visualization_args(parser: argparse.ArgumentParser) -> None:
+    visual = parser.add_mutually_exclusive_group()
+    visual.add_argument(
+        "--visualize",
+        action="store_true",
+        help=(
+            "Request task-specific Gazebo/MoveIt visualization. This fails closed until "
+            "an exact physical Twin and compiled motion plan are available."
+        ),
+    )
+    visual.add_argument(
+        "--visual-demo",
+        action="store_true",
+        help=(
+            "Run the qualified UR5e Gazebo/MoveIt motion demo after engineering verification. "
+            "The demo is visual but is not the BOX/source/destination task-specific motion."
+        ),
+    )
+    parser.add_argument("--ros-setup", default="/opt/ros/jazzy/setup.bash")
+    parser.add_argument("--workspace-setup", default=None)
+    parser.add_argument("--visual-log-dir", default="~/.devagent/ai-visual")
+    parser.add_argument("--startup-timeout", type=float, default=120.0)
+    parser.add_argument("--motion-timeout", type=float, default=90.0)
+    parser.add_argument(
+        "--close-after",
+        action="store_true",
+        help="Close Gazebo after a successful visual demo instead of leaving it open",
+    )
+
+
 def _parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(prog="devagent-physical-ai")
     sub = parser.add_subparsers(dest="command", required=True)
@@ -81,6 +112,7 @@ def _parser() -> argparse.ArgumentParser:
         help="Natural language -> validated request -> AI plan/critic -> deterministic model smoke",
     )
     _add_live_request_args(engineer)
+    _add_visualization_args(engineer)
     return parser
 
 
@@ -203,6 +235,104 @@ def _request(args: argparse.Namespace, progress: ProgressCallback | None = None)
     return 3
 
 
+def _task_visualization_block() -> dict:
+    return {
+        "requested": True,
+        "mode": "task_specific",
+        "launched": False,
+        "passed": False,
+        "task_specific": True,
+        "failure_code": "task_visualization_requires_physical_twin_motion_plan",
+        "required_evidence": [
+            "robot_base_pose",
+            "source_pose_and_collision_geometry",
+            "destination_pose_and_collision_geometry",
+            "workpiece_geometry",
+            "tool_tcp_and_collision_geometry",
+            "compiled_physical_motion_plan",
+        ],
+        "detail": (
+            "The verified high-level pick/move/place graph cannot be represented as an exact "
+            "Gazebo/MoveIt motion without measured or user-declared physical scene geometry and "
+            "a compiled PhysicalMotionPlan. DevAgent will not substitute a generic robot motion "
+            "and label it as the requested task."
+        ),
+        "physical_qualification": False,
+        "real_execution_allowed": False,
+    }
+
+
+def _run_visual_demo(
+    args: argparse.Namespace,
+    *,
+    robot: str,
+    progress: ProgressCallback | None,
+) -> dict:
+    if robot != "ur5e":
+        return {
+            "requested": True,
+            "mode": "robot_motion_demo",
+            "launched": False,
+            "passed": False,
+            "task_specific": False,
+            "robot": robot,
+            "failure_code": "visual_demo_robot_not_supported",
+            "detail": "The packaged visual demo currently supports UR5e only.",
+            "physical_qualification": False,
+            "real_execution_allowed": False,
+        }
+    if args.startup_timeout <= 0 or args.motion_timeout <= 0:
+        raise ValueError("visual_timeouts_must_be_positive")
+
+    from .ros2 import SubprocessRosRunner, UR5eLaptopAcceptance
+
+    runner = SubprocessRosRunner(
+        ros_setup=args.ros_setup,
+        workspace_setup=args.workspace_setup,
+    )
+    root = Path(args.visual_log_dir).expanduser().resolve()
+    report_path = root / "latest-ai-visual-demo.json"
+    _emit(progress, "VISUAL", "checking display and ROS/Gazebo/MoveIt readiness")
+    report = UR5eLaptopAcceptance(runner, ros_setup=args.ros_setup).run(
+        log_dir=root,
+        startup_timeout_s=args.startup_timeout,
+        motion_timeout_s=args.motion_timeout,
+        keep_open=not args.close_after,
+    )
+    report.write_json(report_path)
+    if report.passed:
+        _emit(
+            progress,
+            "VISUAL",
+            (
+                "UR5e motion demo passed; simulator closed"
+                if args.close_after
+                else "UR5e motion demo passed; Gazebo/MoveIt left open for inspection"
+            ),
+        )
+    else:
+        failed = next((stage.name for stage in report.stages if not stage.passed), "unknown")
+        _emit(progress, "VISUAL", f"UR5e motion demo blocked at stage={failed}")
+
+    return {
+        "requested": True,
+        "mode": "robot_motion_demo",
+        "launched": any(stage.name == "simulation_launch" and stage.passed for stage in report.stages),
+        "passed": report.passed,
+        "task_specific": False,
+        "robot": "ur5e",
+        "motion_source": "ur_robot_driver/example_move.py",
+        "warning": (
+            "This validates the workstation visual motion stack only. It is not the exact "
+            "BOX/source/destination motion generated from the engineering request."
+        ),
+        "report_path": str(report_path),
+        "report": report.to_dict(),
+        "physical_qualification": False,
+        "real_execution_allowed": False,
+    }
+
+
 def _engineer(args: argparse.Namespace, progress: ProgressCallback | None = None) -> int:
     _emit(progress, "START", f"provider={args.provider} model={args.model}")
     _, runtime, policy = _live_runtime(
@@ -227,6 +357,33 @@ def _engineer(args: argparse.Namespace, progress: ProgressCallback | None = None
     payload = result.to_dict()
     payload["provider_mode"] = "simulation_only_ephemeral_route"
     payload["real_execution_allowed"] = False
+
+    if result.ready_for_twin_simulation and args.visualize:
+        _emit(
+            progress,
+            "VISUAL",
+            "task-specific visualization requested; checking physical motion evidence",
+        )
+        payload["visualization"] = _task_visualization_block()
+        _emit(
+            progress,
+            "VISUAL",
+            "blocked: exact Twin geometry and compiled PhysicalMotionPlan are required",
+        )
+        print(json.dumps(payload, indent=2))
+        return 6
+
+    if result.ready_for_twin_simulation and args.visual_demo:
+        request = result.interpretation.validated
+        robot = "" if request is None else request.robot
+        payload["visualization"] = _run_visual_demo(
+            args,
+            robot=robot,
+            progress=progress,
+        )
+        print(json.dumps(payload, indent=2))
+        return 0 if payload["visualization"]["passed"] else 6
+
     print(json.dumps(payload, indent=2))
     if result.ready_for_twin_simulation:
         return 0

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5/src/devagent_physical_engine.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devagent-physical-engine
-Version: 0.11.4
+Version: 0.11.5
 Summary: Verification-first agentic planning runtime for robotic and industrial automation.
 Author: Tom Ha
 License: All Rights Reserved
@@ -49,11 +49,11 @@ DevAgent Smart Physical Engine adds AI-assisted engineering to existing automati
 
 ## Current version
 
-**v0.11.4 — Beta software / evidence-gated physical runtime**
+**v0.11.5 — Beta software / evidence-gated physical runtime**
 
 The Python software stack is regression-tested on Python 3.11, 3.12, and 3.13 with separate static/coverage, package-install, and runtime dependency security gates. The UR5e Gazebo/MoveIt path now contains production-candidate continuous-clearance, tool-collision, and workpiece-lifecycle capabilities, but those physical-simulation scopes remain **EXPERIMENTAL until the executable target-workstation v0.11 campaign passes**.
 
-v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 adds live terminal progress tracing for AI request, engineering, and provider-qualification workflows. Progress is written to `stderr`; final machine-readable JSON remains isolated on `stdout`.
+v0.11.3 normalized provider-neutral agent schemas at the OpenAI transport boundary so strict Structured Outputs can represent DevAgent optional fields without changing Anthropic/Gemini or deterministic core semantics. v0.11.4 added live terminal progress tracing for AI request, engineering, and provider-qualification workflows. v0.11.5 adds explicit visualization modes: `--visualize` requests exact task-specific visualization and fails closed until an exact physical Twin plus compiled `PhysicalMotionPlan` exist; `--visual-demo` launches the UR5e Gazebo/MoveIt motion demo without misrepresenting that generic motion as the requested BOX/source/destination task.
 
 Real robot execution remains locked. Functional-safety certification and site qualification are not claimed.
 
@@ -66,6 +66,8 @@ Real robot execution remains locked. Functional-safety certification and site qu
 | Planner / Critic / Recovery | Implemented |
 | Natural-language engineering request front door | Implemented |
 | Live terminal stage tracing (`stderr`) | Implemented; `--quiet` supported |
+| Explicit AI visual demo (`--visual-demo`) | Implemented for UR5e workstation visual motion; not task-specific |
+| Exact task visualization (`--visualize`) | Evidence-gated; blocked until exact Twin + compiled physical motion are available |
 | Deterministic compiler / semantic / verification gates | Implemented |
 | Multi-vendor robot profile registry | Implemented |
 | UR5e / FANUC CRX / KUKA KR / ABB IRB abstraction | Implemented; physical qualification is vendor-specific |
@@ -211,7 +213,7 @@ devagent-physical-ai engineer \
   --provider openai --model <model-id>
 ```
 
-Live commands now show stage-by-stage progress while they are running, for example:
+Live commands show stage-by-stage progress while they are running, for example:
 
 ```text
 [DevAgent +   0.00s] START        provider=openai model=<model-id>
@@ -230,6 +232,30 @@ Live commands now show stage-by-stage progress while they are running, for examp
 
 Progress logs are emitted to `stderr`, so piping `stdout` still produces clean JSON. For CI or scripts that do not want terminal progress, add `--quiet`.
 
+### Visualization semantics
+
+`--visualize` means **visualize the exact engineering task**, not merely make a robot move. Therefore it fails closed until the request has enough physical evidence to build the exact scene and a compiled physical trajectory:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visualize
+```
+
+Until the task-specific physical bridge is complete, the result reports `task_visualization_requires_physical_twin_motion_plan` and lists required evidence such as robot/source/destination poses, collision geometry, workpiece/tool geometry, and a compiled `PhysicalMotionPlan`. DevAgent intentionally does not substitute a canned motion and claim it is the requested BOX task.
+
+To verify that the workstation can actually launch Gazebo/MoveIt and visibly move a UR5e now, use the explicit demo mode:
+
+```bash
+devagent-physical-ai engineer \
+  "Use a UR5e to load BOX_101 from conveyor_a to cnc_04. Simulate and verify the plan." \
+  --provider openai --model <model-id> \
+  --visual-demo
+```
+
+The visual demo uses the packaged UR simulation motion smoke, records an acceptance report under `~/.devagent/ai-visual`, and leaves Gazebo/MoveIt open after success unless `--close-after` is supplied. Its JSON explicitly reports `task_specific=false` and `motion_source=ur_robot_driver/example_move.py`.
+
 Equivalent routing supports Anthropic and Gemini. Live AI qualification does not unlock real robot execution.
 
 ## Software production gates

{devagent_physical_engine-0.11.4 → devagent_physical_engine-0.11.5}/tests/test_ai_cli.py

@@ -95,6 +95,116 @@ class AiCliTests(unittest.TestCase):
         self.assertEqual(json.loads(output.getvalue()), {"passed": True})
         self.assertIsNone(qualify.call_args.kwargs["progress"])
 
+    def test_engineer_parser_accepts_task_visualization(self):
+        args = ai_cli._parser().parse_args(
+            [
+                "engineer",
+                "load box",
+                "--provider",
+                "openai",
+                "--model",
+                "model-id",
+                "--visualize",
+            ]
+        )
+        self.assertTrue(args.visualize)
+        self.assertFalse(args.visual_demo)
+
+    def test_task_visualization_fails_closed_without_physical_motion_plan(self):
+        fake_result = SimpleNamespace(
+            ready_for_twin_simulation=True,
+            failure_code=None,
+            interpretation=SimpleNamespace(
+                validated=SimpleNamespace(robot="ur5e"),
+                rejection_code=None,
+            ),
+            to_dict=lambda: {
+                "ready_for_twin_simulation": True,
+                "physical_qualification": False,
+            },
+        )
+        output = io.StringIO()
+        with patch.object(
+            ai_cli,
+            "_live_runtime",
+            return_value=(object(), object(), object()),
+        ), patch.object(ai_cli, "SmartEngineeringAgent") as smart, patch.object(
+            ai_cli, "_run_visual_demo"
+        ) as demo, contextlib.redirect_stdout(output):
+            smart.return_value.run.return_value = fake_result
+            code = ai_cli.main(
+                [
+                    "engineer",
+                    "load box",
+                    "--provider",
+                    "openai",
+                    "--model",
+                    "model-id",
+                    "--visualize",
+                    "--quiet",
+                ]
+            )
+        payload = json.loads(output.getvalue())
+        self.assertEqual(code, 6)
+        self.assertFalse(payload["visualization"]["launched"])
+        self.assertTrue(payload["visualization"]["task_specific"])
+        self.assertEqual(
+            payload["visualization"]["failure_code"],
+            "task_visualization_requires_physical_twin_motion_plan",
+        )
+        demo.assert_not_called()
+
+    def test_visual_demo_is_explicitly_not_task_specific(self):
+        fake_result = SimpleNamespace(
+            ready_for_twin_simulation=True,
+            failure_code=None,
+            interpretation=SimpleNamespace(
+                validated=SimpleNamespace(robot="ur5e"),
+                rejection_code=None,
+            ),
+            to_dict=lambda: {
+                "ready_for_twin_simulation": True,
+                "physical_qualification": False,
+            },
+        )
+        visual = {
+            "requested": True,
+            "mode": "robot_motion_demo",
+            "launched": True,
+            "passed": True,
+            "task_specific": False,
+            "physical_qualification": False,
+            "real_execution_allowed": False,
+        }
+        output = io.StringIO()
+        with patch.object(
+            ai_cli,
+            "_live_runtime",
+            return_value=(object(), object(), object()),
+        ), patch.object(ai_cli, "SmartEngineeringAgent") as smart, patch.object(
+            ai_cli,
+            "_run_visual_demo",
+            return_value=visual,
+        ) as demo, contextlib.redirect_stdout(output):
+            smart.return_value.run.return_value = fake_result
+            code = ai_cli.main(
+                [
+                    "engineer",
+                    "load box",
+                    "--provider",
+                    "openai",
+                    "--model",
+                    "model-id",
+                    "--visual-demo",
+                    "--quiet",
+                ]
+            )
+        payload = json.loads(output.getvalue())
+        self.assertEqual(code, 0)
+        self.assertTrue(payload["visualization"]["launched"])
+        self.assertFalse(payload["visualization"]["task_specific"])
+        demo.assert_called_once()
+
 
 if __name__ == "__main__":
     unittest.main()