@vention/vention-skills 0.10.27 → 0.10.29

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "vention",
-  "version": "0.10.27",
+  "version": "0.10.29",
   "description": "Vention skills for coding agents: write MachineLogic applications, build machine designs, and diagnose deployed machines.",
   "author": {
     "name": "Vention"

package/README.md

@@ -1,6 +1,6 @@
 # Vention skills
 
-Version 0.10.27. Skills for coding agents that work with Vention machines, packaged as one plugin, `vention`, in the `vention-skills` marketplace.
+Version 0.10.29. Skills for coding agents that work with Vention machines, packaged as one plugin, `vention`, in the `vention-skills` marketplace.
 
 | Skill                       | What it does                                                                                                                         |
 | --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------ |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vention/vention-skills",
-  "version": "0.10.27",
+  "version": "0.10.29",
   "description": "Vention skills for coding agents: write MachineLogic applications, build machine designs, and diagnose deployed machines.",
   "license": "MIT",
   "type": "module",

package/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "vention",
-  "version": "0.10.27",
+  "version": "0.10.29",
   "description": "Vention skills for coding agents: write MachineLogic applications, build machine designs, and diagnose deployed machines.",
   "author": {
     "name": "Vention"

package/skills/vention-design/SKILL.md

@@ -3,7 +3,7 @@ name: vention-design
 description: Create, build and edit Vention machine designs and design graphs for agentic layout. Use when the user wants to start a new design, lay out an automation cell, add or reposition components in a Vention design, generate a machine layout from a description, add or place a floor plan image under a design, list a design's versions, or inspect an existing design graph. Trigger on MachineBuilder designs, design graphs, assembly graphs, extrusions, actuators, robot mounting, cell footprints, floor plans, or a machine-builder URL.
 license: MIT
 metadata:
-  version: "0.10.27"
+  version: "0.10.29"
   author: Vention
 ---
 

package/skills/vention-machine-logic/SKILL.md

@@ -3,7 +3,7 @@ name: vention-machine-logic
 description: Write, edit, validate, and deploy MachineLogic Python applications for Vention machines with the machine-logic-sdk and the Vention CLI. Load it before writing or changing any Python that talks to a Vention machine, however short, since device names and version-specific SDK calls come from here. Use to program a Vention machine or MachineMotion controller, drive actuators, conveyors, robots, pneumatics, or digital I/O, change an existing application, build a state machine, debug from logs, or pull, push, or link with the Vention CLI (vention or vn). Trigger on MachineLogic, machine-logic-sdk, MachineMotion, MachineBuilder applications, the Vention CLI, or a directory holding .machine-code-app-directory-info.json. Printing what the machine or robot holds now (positions, taught targets, calibration, I/O) is a program: files on disk hold only as-designed values. Not for explaining an existing machine or its configuration in words when no program is asked for, nor for Python that never touches the machine.
 license: MIT
 metadata:
-  version: "0.10.27"
+  version: "0.10.29"
   machine-logic-sdk: ">=3.0,<4"
   author: Vention
 ---
@@ -175,8 +175,9 @@ async def feed_board(conveyor: IActuator, line: LineState, travel_mm: float) ->
     line.boards_out += 1                                            # quick calls and shared state stay on the loop
 ```
 
-A blocking call with no `_async` version, such as `home`, goes through `asyncio.to_thread` too; home once, before the tasks start.
+Any other blocking call goes through `asyncio.to_thread` too. Home once, before the tasks start: `await asyncio.to_thread(axis.home, 30)`, or, on SDK releases after 3.1.0, `axis.home_async()` followed by the same wait, which leaves the axis free for a fault path to `stop()` mid-home.
 
+- An actuator's `state.position`, `state.speed` and `state.output_torque` each make a blocking HTTP request to the controller. Inside `async def`, read them with `await asyncio.to_thread(lambda: axis.state.position)`.
 - Start the tasks with `asyncio.gather` or `asyncio.wait`; the target runs Python 3.10, which has no `TaskGroup`.
 - SDK callbacks (`on_state_change`, `machine.on_system_state_change`, MQTT) run on SDK threads. Hand their work to the loop with `loop.call_soon_threadsafe`, never by touching shared state from the callback.
 - Cancelling a task does not stop a move a worker thread is waiting on. The shutdown path stops every device it started, in a `finally`.

package/skills/vention-machine-logic-hmi/SKILL.md

@@ -3,7 +3,7 @@ name: vention-machine-logic-hmi
 description: Build or edit the custom operator HMI of a MachineLogic application, the React touch screen in customui/ that runs in MachineBuilder and on the MachineMotion pendant, wired to the application's Python process through vention-communication. Use when the user asks for an operator screen, a custom HMI or UI, a touch panel, start and stop buttons, live machine state on screen, or recipes an operator picks, for a Vention machine or inside a directory holding .machine-code-app-directory-info.json. Uses @vention/machine-ui, @vention/machine-apps-components and @vention/machine-logic-ui-sdk. Do not use for the no-code HMI Builder, or for the Python motion logic alone (that is vention-machine-logic).
 license: MIT
 metadata:
-  version: "0.10.27"
+  version: "0.10.29"
   author: Vention
 ---
 
@@ -66,15 +66,15 @@ async def start(request: StartRequest) -> CommandResponse:
 
 
 app.register_rpc_plugin(build_state_machine_rpc_bundle(station, triggers=[Triggers.stop.name, BaseTriggers.RESET.value]))
-app.register_rpc_plugin(build_storage_rpc_bundle(accessors=[recipe_accessor], max_records_per_model=None))
+app.register_rpc_plugin(build_storage_rpc_bundle(accessors=[recipe_accessor], max_records_per_model=None, enable_db_actions=False))
 app.finalize()
 ```
 
 - Create `Machine()` in the FastAPI `lifespan`, through `asyncio.to_thread`, never at import. Importing `server.py` must work with no machine attached: that is how the proto is emitted. When it fails, show the failure as a fault the operator can read.
 - `@action("Start")` names the RPC `Start`. Without the name the RPC takes the function's name as written.
-- The state machine bundle adds `GetState`, `GetHistory` and one `Trigger_<Name>` per trigger you list. List only the triggers the operator may fire.
+- The state machine bundle adds `GetState`, `GetHistory` and one `Trigger_<Name>` per trigger. Always pass `triggers=` with only the ones the operator may fire: without it the bundle exposes every trigger, `to_fault` included, and the HMI can fire them around the checks your `@action` handlers make.
 - The storage bundle adds `Recipe_ListRecords`, `Recipe_CreateRecord`, `Recipe_UpdateRecord` and `Recipe_DeleteRecord` among others, named after the accessor. `max_records_per_model=None` lifts its default cap of 5 records.
-- The storage bundle also adds `Database_*` RPCs, including a whole-database restore. Never expose them to operators.
+- Pass `enable_db_actions=False` unless the HMI has a backup or restore screen. Otherwise the bundle adds `Database_*` RPCs the app never uses, including a whole-database restore.
 - Give every storage model field a zero-value default (`dwell_ms: int = 0`). The client leaves zero values out of the JSON it sends, and a field with no default then fails as NULL.
 - Every storage write takes a non-empty `actor`, the name recorded in the audit log.
 
@@ -101,7 +101,7 @@ async def publish_event(level: str, message: str, detail: str) -> Event:
     return Event(timestamp=datetime.now(timezone.utc).isoformat(), level=level, message=message, detail=detail)
 ```
 
-Awaiting the function publishes its result. Publish status from an `@on_state_change` method on the state machine (`self.spawn(publish_status())`) and after every count or fault change. `replay=True` hands a new subscriber the last status at once; the event log uses `replay=False` so a reconnect does not repeat old events. Also republish status every `STATUS_HEARTBEAT_S` from a task the `lifespan` starts, as the example does: a connection can hang without closing, and the HMI only notices when the heartbeat stops.
+Awaiting the function publishes its result. Publish status from an `@on_state_change` method on the state machine (`self.spawn(publish_status())`) and after every count or fault change. `replay=True` hands a new subscriber the last status at once; the event log uses `replay=False` so a reconnect does not repeat old events. Keep the last events in a `deque(maxlen=RECENT_EVENT_COUNT)` that `publish_event` appends to and an `@action("RecentEvents")` returns: the HMI loads them each time it connects, so the Logs page is not empty after a reload. Also republish status every `STATUS_HEARTBEAT_S` from a task the `lifespan` starts, as the example does: a connection can hang without closing, and the HMI only notices when the heartbeat stops.
 
 ## The generated client
 
@@ -190,7 +190,7 @@ The HMI drives physical machinery that can injure someone.
 
 `examples/indexing-station-hmi/` indexes a part on a conveyor, clamps it, proves the clamp closed, dwells and releases, for the cycle count of the recipe the operator picked.
 
-- `server.py`: the state machine, the streams, the Start action, both bundles, `Machine()` in the `lifespan`, the emergency stop routed to fault, and every device stopped on shutdown.
+- `server.py`: the state machine, the streams and recent-event buffer, the Start and RecentEvents actions, both bundles, `Machine()` in the `lifespan`, the emergency stop routed to fault, and every device stopped on shutdown.
 - `models.py`: the `Recipe` storage model, every field with a zero default.
 - `project.json`, `requirements.txt`: the one process and the exact pins.
 - `proto/app.proto`: what `finalize()` emits.

package/skills/vention-machine-logic-hmi/examples/indexing-station-hmi/customui/package.json

@@ -15,9 +15,9 @@
     "@emotion/styled": "11.14.1",
     "@mui/material": "5.16.7",
     "@tabler/icons-react": "3.44.0",
-    "@vention/machine-apps-components": "1.30.25",
-    "@vention/machine-logic-ui-sdk": "0.47.4",
-    "@vention/machine-ui": "6.7.0",
+    "@vention/machine-apps-components": "1.30.26",
+    "@vention/machine-logic-ui-sdk": "0.47.5",
+    "@vention/machine-ui": "6.7.1",
     "react": "17.0.2",
     "react-dom": "17.0.2",
     "react-router-dom": "6.30.3",

package/skills/vention-machine-logic-hmi/examples/indexing-station-hmi/customui/src/app.tsx

@@ -28,29 +28,68 @@ function toLogLevel(level: string): LogType {
   return "info"
 }
 
+function eventKey(entry: LogEntry): string {
+  return `${entry.date}|${entry.message}|${entry.description}`
+}
+
+// The stream and RecentEvents can both deliver an event published while the HMI connects, in either order.
+function mergeNewestFirst(previous: LogEntry[], incoming: LogEntry[]): LogEntry[] {
+  const shown = new Set(previous.map(eventKey))
+  const merged = [...previous, ...incoming.filter(entry => !shown.has(eventKey(entry)))]
+  return merged.sort((left, right) => right.date.localeCompare(left.date)).slice(0, MAX_LOG_ENTRIES)
+}
+
 /** The server's events, plus the HMI's own command failures, newest first. */
-function useEventLog(): { entries: LogEntry[]; addEntry: (level: LogType, message: string, description: string) => string } {
+function useEventLog(connected: boolean): {
+  entries: LogEntry[]
+  addEntry: (level: LogType, message: string, description: string) => string
+} {
   const { value: latestEvent } = useStream<Event>(openEvents)
   const [entries, setEntries] = useState<LogEntry[]>([])
   const nextId = useRef(0)
 
-  const addEntry = useCallback((level: LogType, message: string, description: string, date = new Date().toISOString()) => {
+  const toLogEntry = useCallback((level: LogType, message: string, description: string, date: string): LogEntry => {
     nextId.current += 1
-    const entry: LogEntry = { id: String(nextId.current), date, level, code: level.toUpperCase(), message, description }
-    setEntries(previous => [entry, ...previous].slice(0, MAX_LOG_ENTRIES))
-    return entry.id
+    return { id: String(nextId.current), date, level, code: level.toUpperCase(), message, description }
   }, [])
 
+  const addEntry = useCallback(
+    (level: LogType, message: string, description: string, date = new Date().toISOString()) => {
+      const entry = toLogEntry(level, message, description, date)
+      setEntries(previous => mergeNewestFirst(previous, [entry]))
+      return entry.id
+    },
+    [toLogEntry]
+  )
+
   useEffect(() => {
     if (latestEvent !== undefined) addEntry(toLogLevel(latestEvent.level), latestEvent.message, latestEvent.detail, latestEvent.timestamp)
   }, [latestEvent, addEntry])
 
+  // The Events stream does not replay, so each (re)connect loads what the server kept.
+  useEffect(() => {
+    if (!connected) return
+    const controller = new AbortController()
+    client
+      .recentEvents({}, { signal: controller.signal })
+      .then(({ events }) =>
+        setEntries(previous =>
+          mergeNewestFirst(
+            previous,
+            events.map(event => toLogEntry(toLogLevel(event.level), event.message, event.detail, event.timestamp))
+          )
+        )
+      )
+      .catch(() => undefined)
+    return () => controller.abort()
+  }, [connected, toLogEntry])
+
   return { entries, addEntry }
 }
 
 export function App() {
   const { value: status, connected } = useStream(openStatus, STATUS_STALL_MS)
-  const { entries: logEntries, addEntry } = useEventLog()
+  const { entries: logEntries, addEntry } = useEventLog(connected)
   const navigate = useNavigate()
   const isOnLogsPage = useLocation().pathname === "/logs"
   // The latest error shows as a banner on every page until it fades; View opens the whole entry on the Logs page.

package/skills/vention-machine-logic-hmi/examples/indexing-station-hmi/proto/app.proto

@@ -4,6 +4,17 @@ package vention.app.v1;
 import "google/protobuf/empty.proto";
 
 
+message Event {
+  string timestamp = 1;
+  string level = 2;
+  string message = 3;
+  string detail = 4;
+}
+
+message EventHistory {
+  repeated Event events = 1;
+}
+
 message StartRequest {
   int32 recipe_id = 1;
 }
@@ -117,42 +128,6 @@ message RecipeRestoreResponse {
   Recipe record = 1;
 }
 
-message HealthResponse {
-  string status = 1;
-}
-
-message AuditQueryRequest {
-  string component = 1;
-  int32 record_id = 2;
-  string actor = 3;
-  string operation = 4;
-  string since = 5;
-  string until = 6;
-  int32 limit = 7;
-  int32 offset = 8;
-}
-
-message AuditListResponse {
-  repeated string records = 1;
-}
-
-message FileResponse {
-  string filename = 1;
-  bytes bytes = 2;
-}
-
-message DatabaseRestoreRequest {
-  bytes bytes = 1;
-  string filename = 2;
-  bool integrity_check = 3;
-  bool dry_run = 4;
-}
-
-message DatabaseRestoreResponse {
-  bool ok = 1;
-  string message = 2;
-}
-
 message StationStatus {
   string state = 1;
   string recipe_name = 2;
@@ -165,14 +140,8 @@ message StationStatus {
   string fault = 9;
 }
 
-message Event {
-  string timestamp = 1;
-  string level = 2;
-  string message = 3;
-  string detail = 4;
-}
-
 service IndexingStationService {
+  rpc RecentEvents (google.protobuf.Empty) returns (EventHistory);
   rpc Start (StartRequest) returns (CommandResponse);
   rpc GetState (google.protobuf.Empty) returns (StateResponse);
   rpc GetHistory (google.protobuf.Empty) returns (HistoryResponse);
@@ -185,11 +154,6 @@ service IndexingStationService {
   rpc Recipe_UpdateRecord (RecipeUpdateRequest) returns (RecipeUpdateResponse);
   rpc Recipe_DeleteRecord (CrudDeleteRequest) returns (RecipeStatusResponse);
   rpc Recipe_RestoreRecord (CrudRestoreRequest) returns (RecipeRestoreResponse);
-  rpc Database_Health (google.protobuf.Empty) returns (HealthResponse);
-  rpc Database_ReadAudit (AuditQueryRequest) returns (AuditListResponse);
-  rpc Database_ExportZip (google.protobuf.Empty) returns (FileResponse);
-  rpc Database_BackupSqlite (google.protobuf.Empty) returns (FileResponse);
-  rpc Database_RestoreSqlite (DatabaseRestoreRequest) returns (DatabaseRestoreResponse);
   rpc Status (google.protobuf.Empty) returns (stream StationStatus);
   rpc Events (google.protobuf.Empty) returns (stream Event);
 }

package/skills/vention-machine-logic-hmi/examples/indexing-station-hmi/requirements.txt

@@ -1,5 +1,5 @@
 machine-logic-sdk==3.1.0
-vention-communication==0.15.35
-vention-state-machine==0.16.29
-vention-storage==0.20.13
+vention-communication==0.15.36
+vention-state-machine==0.16.31
+vention-storage==0.20.14
 uvicorn==0.35.0

package/skills/vention-machine-logic-hmi/examples/indexing-station-hmi/server.py

@@ -6,6 +6,7 @@ An emergency stop sends the station to fault, and shutting the server down stops
 
 import asyncio
 import time
+from collections import deque
 from contextlib import asynccontextmanager
 from datetime import datetime, timezone
 from pathlib import Path
@@ -33,6 +34,8 @@ CLAMP_PROOF_TIMEOUT_S = 2.0
 CLAMP_POLL_S = 0.05
 # The HMI treats a Status stream that stays silent for three heartbeats as dropped.
 STATUS_HEARTBEAT_S = 2.0
+# The Events stream does not replay, so the HMI loads these each time it connects.
+RECENT_EVENT_COUNT = 100
 ACCELERATION_MM_S2 = 400.0
 DATA_DIR = Path(__file__).parent / "data"
 
@@ -79,6 +82,10 @@ class Event(BaseModel):
     detail: str
 
 
+class EventHistory(BaseModel):
+    events: list[Event]
+
+
 class StartRequest(BaseModel):
     recipe_id: int
 
@@ -224,6 +231,7 @@ DATA_DIR.mkdir(exist_ok=True)
 bootstrap(database_url=f"sqlite:///{DATA_DIR / 'recipes.db'}")
 recipe_accessor = ModelAccessor(Recipe, "recipe")
 station = IndexingStation()
+recent_events: deque[Event] = deque(maxlen=RECENT_EVENT_COUNT)
 
 
 async def repeat_status() -> None:
@@ -266,7 +274,14 @@ async def publish_status() -> StationStatus:
 
 @stream(name="Events", payload=Event, replay=False, queue_maxsize=100, policy="fifo")
 async def publish_event(level: str, message: str, detail: str) -> Event:
-    return Event(timestamp=datetime.now(timezone.utc).isoformat(), level=level, message=message, detail=detail)
+    event = Event(timestamp=datetime.now(timezone.utc).isoformat(), level=level, message=message, detail=detail)
+    recent_events.append(event)
+    return event
+
+
+@action("RecentEvents")
+async def get_recent_events() -> EventHistory:
+    return EventHistory(events=list(recent_events))
 
 
 @action("Start")
@@ -286,5 +301,5 @@ async def start(request: StartRequest) -> CommandResponse:
 
 
 app.register_rpc_plugin(build_state_machine_rpc_bundle(station, triggers=[Triggers.stop.name, BaseTriggers.RESET.value]))
-app.register_rpc_plugin(build_storage_rpc_bundle(accessors=[recipe_accessor], max_records_per_model=None))
+app.register_rpc_plugin(build_storage_rpc_bundle(accessors=[recipe_accessor], max_records_per_model=None, enable_db_actions=False))
 app.finalize()

package/skills/vention-monitoring/SKILL.md

@@ -3,7 +3,7 @@ name: vention-monitoring
 description: Diagnose deployed Vention machines from controller state, history, health, and activity data. Use when the user reports a machine that is down, faulted, offline, slow, or underperforming, asks why a cycle time regressed, wants to interpret MachineMotion event data or drive and brake status, asks about OEE or uptime, or wants to know which machine in a fleet needs attention. Trigger on a MachineMotion serial number plus any complaint.
 license: MIT
 metadata:
-  version: "0.10.27"
+  version: "0.10.29"
   author: Vention
 ---