菜单
Read the plant-floor Azure IoT surface — device twins (spindles, conveyors, robotic arms, vision systems), time-series telemetry with pre-annotated alarms, downtime events with cause codes and lot ids, and per-line OEE rollups (Availability × Performance × Quality).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Share generated files (PDFs, charts, CSVs) with Frank by writing them to /tmp and referencing the absolute path. The frontend auto-converts /tmp paths into download links.
Build the downloadable incident brief PDF — cover + KPI strip + OEE chart + downtime timeline + root-cause narrative + supplier/material signals + recommended actions + sign-off, per SOP §7.1. Renders via the inline HTML template + Playwright.
Microsoft 365 surface — READ Frank's mailbox / Wes's calendar / OneDrive (shift reports, SOPs) AND a strict H-I-T-L workflow to draft and send an email (e.g. brief Wes on the morning incident).
Compute and plot OEE — a 7-day trend chart with target/threshold lines, an overlay of telemetry alarms and downtime stops, plus a flat CSV export. Saves /tmp/oee-trend-<plant>-line-<N>-<date>.png + .csv.
The Olympus Cleveland-plant Standard Operating Procedure — alarm thresholds, escalation matrix, supplier quality-hold rules, OEE definitions, incident-brief template requirements. Cite the section number when answering.
Read SAP S/4HANA manufacturing data — production orders, materials, vendors, plants, lines. Pivot here once the IoT signal names a `related_po_id` to find the order, the material, and the (possibly QA-blocked) vendor.
| name | azure-iot |
| description | Read the plant-floor Azure IoT surface — device twins (spindles, conveyors, robotic arms, vision systems), time-series telemetry with pre-annotated alarms, downtime events with cause codes and lot ids, and per-line OEE rollups (Availability × Performance × Quality). |
| enabled | true |
This is the read interface to azure-iot-mcp-server. This is where every Line-N investigation starts — OEE tells you which line is missing target; downtime events tell you which device and why; telemetry tells you the underlying signal.
| User intent | Tool |
|---|---|
| List the equipment on the floor (filter by plant, line, status) | iot_list_devices |
| One device with full detail (vendor/model/firmware, alarm thresholds, status_note) | iot_get_device |
Pull time-series readings for a device (vibration, temp, load, RPM, etc.) with pre-annotated alarms array | iot_get_telemetry |
List downtime events with cause codes (VIB_ALARM, MATERIAL_HOLD_QA, CAL_DRIFT, PLANNED_MAINT), lot ids, and related_po_id for cross-MCP joins | iot_list_downtime_events |
Per-line per-day OEE with vs_target_pct and flag (on_target / watch / investigate) | iot_get_oee |
The IoT story is OEE → downtime → telemetry → device twin. Walk that order; don't query everything up-front.
iot_get_oee(plant_id="P-CLE") → which line is investigate? Look at the 7-day trend, not just today.iot_list_downtime_events(plant_id="P-CLE", line="<that line>", since="<yesterday>") → what stopped it? Read the cause_code, lot_id, related_po_id.iot_get_device("<the device named in the events>") → is the device twin in Warning / Degraded? Read the status_note.iot_get_telemetry("<device>", since="<window>") → confirm the signal. The alarms array on each reading already lists the breaches — don't re-derive thresholds in your head.related_po_id or a Northbridge-style lot id, pivot to sap-s4; if you need to log a maintenance ticket, pivot to servicenow.alarms array is the source of truth on threshold breaches. Quote it: "vibration crossed 4.5 mm/s at 22:00 — reading was 4.7". Don't compute it.P-CLE here = P-CLE there. Lines too (Line 3 — Precision is the same string).DEV-* map to servicenow CMDB CI-DEV-* by numeric suffix. DEV-3001 → CI-DEV-3001. The servicenow CI's owner_group is Plant Maintenance — that's the assignment group for the work order.on_target / watch / investigate) — use those words; don't invent synonyms.-04:00, Munich +02:00). When Frank says "last shift" assume the prior 8 h ending at the current time.Prefer compact tables — one row per device, one row per downtime event, one row per day for OEE. Lead each row with the id, the metric, and the flag.
code_interpreter to retype numbers — pipe them through that skill)