test(config): cover utils.* node reconciliation as match (CHE-656)

.env.example

@@ -12,6 +12,20 @@ ANTHROPIC_API_KEY=
 OPEN_ROUTER_API_KEY=
 XAI_API_KEY=
 
+# Optional provider-compatible endpoints. OpenAI-compatible URLs include `/v1`;
+# Anthropic URLs are the API origin because the SDK appends `/v1`. Gemini and
+# Vertex values override the SDK service endpoint. Per-model `api_base` in
+# artemis.jsonc overrides these values.
+# GOOGLE_BASE_URL=https://generativelanguage.googleapis.com
+# VERTEX_AI_BASE_URL=https://us-central1-aiplatform.googleapis.com
+# OPENAI_BASE_URL=https://openai-proxy.example/v1
+# ANTHROPIC_BASE_URL=https://anthropic-proxy.example
+# OPEN_ROUTER_BASE_URL=https://openrouter.ai/api/v1
+# XAI_BASE_URL=https://api.x.ai/v1
+# OLLAMA_BASE_URL=http://localhost:11434/v1
+# VLLM_BASE_URL=http://localhost:8000/v1
+# CUSTOM_BASE_URL=http://localhost:8000/v1
+
 # Google Cloud Vision OCR (Optional - for advanced OCR processing)
 OCR_API_KEY=
 

.github/workflows/ci.yml

@@ -17,6 +17,8 @@
   UV_NO_PROGRESS: "1"
 
 jobs:
+  # Windows leg removed org-wide per CHE-522 (Cheese: Linux CI only, no
+  # Windows jobs or artifacts).
   python-quality:
     name: Python 3.12 / ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
@@ -25,15 +27,14 @@
       fail-fast: false
       matrix:
         os:
-          - ubuntu-latest
-          - windows-latest
+          - blacksmith-2vcpu-ubuntu-2404
 
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
 
       - name: Set up Python 3.12
         uses: actions/setup-python@v5
         with:
           python-version: "3.12"
 
@@ -41,7 +42,7 @@
         run: python scripts/check_dependency_sources.py
 
       - name: Set up uv
         uses: astral-sh/setup-uv@v6
         with:
           enable-cache: true
           cache-dependency-glob: uv.lock
@@ -78,7 +79,7 @@
 
   frontend:
     name: Frontend tests and production build
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     timeout-minutes: 20
     defaults:
       run:
@@ -86,10 +87,10 @@
 
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
 
       - name: Set up Node.js
         uses: actions/setup-node@v4
         with:
           node-version: "22.22.3"
           cache: npm
@@ -106,26 +107,26 @@
 
   package:
     name: Build and smoke-test Python packages
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     timeout-minutes: 30
 
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
 
       - name: Set up Python 3.12
         uses: actions/setup-python@v5
         with:
           python-version: "3.12"
 
       - name: Set up uv
         uses: astral-sh/setup-uv@v6
         with:
           enable-cache: true
           cache-dependency-glob: uv.lock
 
       - name: Set up Node.js
         uses: actions/setup-node@v4
         with:
           node-version: "22.22.3"
           cache: npm
@@ -171,7 +172,7 @@
           artemis --help
 
       - name: Upload distributions
         uses: actions/upload-artifact@v4
         with:
           name: python-distributions
           path: dist/*

.quality-baseline.json

@@ -1,5 +1,5 @@
 {
-  "broad_exception_handlers": 754,
+  "broad_exception_handlers": 759,
   "silent_broad_exception_handlers": 0,
   "type_ignore_comments": 18
 }

apps/admin_console/database/repositories/session_repository.py

@@ -510,6 +510,40 @@ def get_session_status(self, session_id: str) -> str | None:
         except Exception:
             return None
 
+    def create_queued_session(
+        self,
+        session_id: str,
+        goal: str,
+        profile: str,
+        device_serial: str | None,
+        start_time: float | None = None,
+    ) -> bool:
+        """Persist a queue item before its worker starts a session."""
+        try:
+            with db_session(self.db_path) as conn:
+                cursor = conn.cursor()
+                cursor.execute(
+                    "INSERT INTO sessions "
+                    "(session_id, initial_goal, start_time, end_time, status, device_info, pid) "
+                    "VALUES (?, ?, ?, NULL, 'queued', ?, NULL)",
+                    (
+                        str(session_id),
+                        goal,
+                        start_time if start_time is not None else time.time(),
+                        json.dumps(
+                            {
+                                "profile": profile,
+                                "device_id": device_serial,
+                            }
+                        ),
+                    ),
+                )
+                conn.commit()
+                return cursor.rowcount > 0
+        except Exception:
+            logger.exception("Could not create queued session %s", session_id)
+            return False
+
     def update_session_status(
         self, session_id: str, status: str, end_time: float | None = None
     ) -> bool:

apps/admin_console/routers/system.py

@@ -422,9 +422,12 @@ def mask_key(k: str | None) -> str | None:
     anthropic_real = get_real_env_value("ANTHROPIC_API_KEY", "anthropic")
     openrouter_real = get_real_env_value("OPEN_ROUTER_API_KEY", "openrouter")
     xai_real = get_real_env_value("XAI_API_KEY", "xai")
-    base_url_val = settings.OPENAI_BASE_URL or os.environ.get("OPENAI_BASE_URL")
-    if base_url_val and is_placeholder_key(base_url_val):
-        base_url_val = None
+    openai_base_url = settings.OPENAI_BASE_URL or os.environ.get("OPENAI_BASE_URL")
+    anthropic_base_url = settings.ANTHROPIC_BASE_URL or os.environ.get("ANTHROPIC_BASE_URL")
+    if openai_base_url and is_placeholder_key(openai_base_url):
+        openai_base_url = None
+    if anthropic_base_url and is_placeholder_key(anthropic_base_url):
+        anthropic_base_url = None
     ocr_real = get_real_env_value("OCR_API_KEY", "ocr") or get_real_env_value(
         "VISION_API_KEY", "ocr"
     )
@@ -467,10 +470,17 @@ def mask_key(k: str | None) -> str | None:
         },
         {
             "name": "OPENAI_BASE_URL",
-            "provider": "custom",
-            "is_set": bool(base_url_val),
-            "preview": base_url_val,
-            "description": "Custom API endpoint (for local Ollama, vLLM, DeepSeek, or proxies)",
+            "provider": "openai",
+            "is_set": bool(openai_base_url),
+            "preview": openai_base_url,
+            "description": "Custom OpenAI-compatible API endpoint",
+        },
+        {
+            "name": "ANTHROPIC_BASE_URL",
+            "provider": "anthropic",
+            "is_set": bool(anthropic_base_url),
+            "preview": anthropic_base_url,
+            "description": "Custom Anthropic-compatible API endpoint",
         },
         {
             "name": "VISION_API_KEY",

apps/admin_console/routers/tasks.py

@@ -178,6 +178,7 @@ async def run_task(request: RunRequest):
         ingress=request.ingress or "frontend",
         session_id=request.session_id,
         conversation_id=request.conversation_id,
+        run_id=request.run_id,
     )
 
 

apps/admin_console/schemas/task_schema.py

@@ -32,6 +32,7 @@ class RunRequest(BaseModel):
     ingress: str | None = "frontend"
     session_id: str | None = None
     conversation_id: str | None = None
+    run_id: str | None = None
 
 
 class ReplayRequest(BaseModel):

apps/admin_console/services/task_queue_service.py

@@ -507,6 +507,7 @@ def _build_worker_invocation(
         explorer_mode = task_item.get("explorer_mode")
         locked_app = task_item.get("locked_app_package") or task_item.get("locked_app")
         app_path = task_item.get("app_path")
+        run_id = task_item.get("run_id")
 
         test_name = f"web_{int(time.time())}_{run_key[:8]}"
         env = os.environ.copy()
@@ -543,6 +544,8 @@ def _build_worker_invocation(
         ]
         if sess_id:
             cmd.extend(["--session-id", str(sess_id)])
+        if run_id:
+            cmd.extend(["--run-id", str(run_id)])
         if expected_output:
             cmd.extend(["--output-description", str(expected_output)])
         if enable_outputter is not None:
@@ -1004,6 +1007,7 @@ def _create_queue_item(
         conversation_id: str | None,
         verification_level: str | None = None,
         explorer_mode: str | None = None,
+        run_id: str | None = None,
     ) -> dict[str, Any]:
         """Reserve a device slot and build one pending queue item for a goal."""
         sess_id = single_session_id if single_session_id else str(uuid.uuid4())
@@ -1031,6 +1035,7 @@ def _create_queue_item(
             "adb_endpoint": endpoint.to_dict(),
             "ingress": ingress,
             "conversation_id": conversation_id,
+            "run_id": run_id,
             "status": "pending",
             "queue_ticket": queue_ticket,
             "created_at": now + index * 0.001,
@@ -1052,12 +1057,18 @@ async def enqueue_tasks(
         conversation_id: str | None = None,
         verification_level: str | None = None,
         explorer_mode: str | None = None,
+        run_id: str | None = None,
     ) -> dict[str, Any]:
         """Enqueues one or more goals and wakes up the background worker.
 
         ``verification_level`` and ``explorer_mode`` are Pro-profile tuning knobs
         forwarded to the worker as ``--verification-level`` / ``--explorer-pro-mode``;
         they are normalised here so the queue item and the CLI see one spelling.
+
+        ``run_id`` is the Gate 1 batch-grouping key (see
+        ``artemis.config.attempt_lifecycle_hooks``); it is forwarded to the
+        spawned worker as ``--run-id`` so daemon-dispatched attempts get the
+        same manifest/reconciliation evidence as standalone runs.
         """
         verification_level = (
             str(verification_level).strip().lower() or None if verification_level else None
@@ -1066,6 +1077,7 @@ async def enqueue_tasks(
         cls.ensure_worker_running()
 
         enqueued_tasks = []
+        created_trace_session_ids: set[str] = set()
         now = time.time()
         endpoint = current_adb_endpoint()
 
@@ -1105,7 +1117,64 @@ async def enqueue_tasks(
                 conversation_id,
                 verification_level=verification_level,
                 explorer_mode=explorer_mode,
+                run_id=run_id,
+            )
+            session_id = str(task_item["session_id"])
+            existing_trace = trace_store.read_status(session_id)
+            trace_created = existing_trace is None
+            existing_trace_is_terminal = bool(
+                existing_trace
+                and existing_trace.get("status") in {"completed", "failed", "cancelled", "success"}
             )
+            try:
+                if existing_trace_is_terminal:
+                    raise RuntimeError(f"Session {session_id} already has a terminal trace")
+                if trace_created:
+                    trace_store.init_trace(
+                        session_id,
+                        goal,
+                        task_item["profile"],
+                        task_item.get("conversation_id"),
+                        task_item.get("device_serial"),
+                    )
+                    created_trace_session_ids.add(session_id)
+                if not session_repo.create_queued_session(
+                    session_id,
+                    goal,
+                    task_item["profile"],
+                    task_item.get("device_serial"),
+                    task_item.get("start_time"),
+                ):
+                    raise RuntimeError(f"Could not persist queued session {session_id}")
+            except (OSError, RuntimeError) as exc:
+                DeviceExecutionLock.cancel_reservation(task_item.get("queue_ticket"))
+                if trace_created or (
+                    not existing_trace_is_terminal
+                    and session_repo.get_session_by_id(session_id) is None
+                ):
+                    try:
+                        trace_store.update_trace_status(session_id, "failed", error=str(exc))
+                    except OSError:
+                        logger.exception(
+                            "Could not mark queue setup failure for session %s", session_id
+                        )
+                for enqueued_task in enqueued_tasks:
+                    enqueued_session_id = str(enqueued_task["session_id"])
+                    session_repo.update_session_status(enqueued_session_id, "failed", time.time())
+                    if enqueued_session_id in created_trace_session_ids:
+                        try:
+                            trace_store.update_trace_status(
+                                enqueued_session_id,
+                                "failed",
+                                error="Task batch could not be queued.",
+                            )
+                        except OSError:
+                            logger.exception(
+                                "Could not mark rolled-back queue session %s failed",
+                                enqueued_session_id,
+                            )
+                    cls._remove_task(enqueued_session_id)
+                raise
             state.queue_items.append(task_item)
             enqueued_tasks.append(task_item)
             cls._broadcast_startup_progress(

artemis/agents/flash/runner.py

@@ -81,9 +81,7 @@
 from artemis.mcp.observation import observe
 from artemis.memory.transcript import PRO_UI_LIST_MARKER, TranscriptLedger, mark_ephemeral
 from artemis.services.llm import (
-    RobustChatModelWrapper,
     acomplete,
-    get_google_llm,
     get_llm,
     invoke_llm_with_timeout_message,
 )
@@ -279,12 +277,7 @@ def _build_ledger(self) -> TranscriptLedger:
 
     def _init_llm(self):
         """Initializes the Universal LLM via the Service Layer."""
-        try:
-            return get_llm(self.ctx, name="operator")
-        except Exception as e:
-            logger.warning(f"Failed to get operator LLM from config, using default: {e}")
-
-            return RobustChatModelWrapper(get_google_llm(model_name="gemini-2.5-flash"), self.ctx)
+        return get_llm(self.ctx, name="operator")
 
     def _render_system_prompt(self, tools_declaration: list) -> str:
         """Renders the system prompt from the flash_runner.md template.

artemis/agents/flash/summarizer.py

@@ -36,7 +36,7 @@
 
 from artemis.context import ArtemisContext
 from artemis.memory.step_memory import JobKey, StepMemoryService
-from artemis.services.llm import RobustChatModelWrapper, get_google_llm, get_llm
+from artemis.services.llm import RobustChatModelWrapper, get_llm
 from artemis.services.token_meter import record_llm_usage
 from artemis.utils.logger import get_logger
 from artemis.utils.task_tree import format_actions_clean
@@ -161,18 +161,22 @@ def __init__(
             flush_timeout_s=flush_timeout_s,
         )
 
-        # Initialize lightweight VLM: prioritize explicit model_name
-        target_model = model_name or "gemini-2.5-flash-lite"
-        self._model_name = target_model
-        try:
-            if model_name:
-                self._llm = get_google_llm(model_name=target_model, temperature=0.0)
-            else:
-                self._llm = get_llm(ctx, name="summarizer", is_utils=True)
-        except Exception:
-            self._llm = get_google_llm(model_name=target_model, temperature=0.0)
+        # The summarizer role carries the provider and default model. This profile
+        # knob only overrides that model when the caller explicitly supplies one.
+        self._model_name = model_name or ""
+        self._llm = get_llm(
+            ctx,
+            name="summarizer",
+            temperature=0.0,
+            model_name=model_name or None,
+        )
         try:
-            configured = getattr(self._llm, "model", None) or getattr(self._llm, "model_name", None)
+            endpoint = getattr(self._llm, "endpoint", None)
+            configured = getattr(endpoint, "model_name", None)
+            if not configured:
+                configured = getattr(self._llm, "model", None) or getattr(
+                    self._llm, "model_name", None
+                )
             if isinstance(configured, str) and configured:
                 self._model_name = configured
         except Exception as exc:

artemis/agents/image_processor/image_processor.py

@@ -79,10 +79,15 @@ def __init__(self, ctx: ArtemisContext):
 
     @trace(type="agent", name="image_processor")
     async def run(self, instruction: str, target_image_path: str) -> dict:
-        try:
-            llm = get_llm(self.ctx, name="image_processor")
-        except Exception:
-            llm = get_llm(self.ctx, name="operator")
+        # There is no ``image_processor`` node on ``LLMConfig`` or
+        # ``LLMConfigUtils``, so the former ``get_llm(name="image_processor")``
+        # attempt raised ``AttributeError`` on every call and a bare handler
+        # silently resolved the operator model instead. The operator model is
+        # what this agent has always actually used; say so rather than hiding
+        # it behind a lookup that cannot succeed. Giving the image processor
+        # its own configurable node is a config-schema change, tracked
+        # separately.
+        llm = get_llm(self.ctx, name="operator")
 
         base_dir = settings.TRACES_PATH
         image_processor_dir = base_dir / "images" / "image_processor"

artemis/agents/object_detector/object_detector.py

@@ -124,8 +124,13 @@ async def _run_object_detection(
     queries = queries or []
     templates = templates or ["Point to the following objects: {labels_str}"]
     try:
-        llm = get_llm(ctx, name="object_detector")
-    except Exception:
+        llm = get_llm(ctx, name="object_detector", is_utils=True)
+    except (ValueError, AttributeError) as exc:
+        logger.warning(
+            f"object_detector is not configured ({exc}); falling back to the operator "
+            "model, which is slower and more expensive. Configure 'object_detector' "
+            "under the llm utils config to use the intended detector."
+        )
         llm = get_llm(ctx, name="operator")
 
     raw_timeout = getattr(getattr(ctx, "llm_config", None), "timeout", None)