メニュー
Daily quality insights for the 3-machine factory line. Combines quality reports (xlsx), machine status (json), production orders (json), and MQTT events to surface root causes for the past 7 days. Emits useful findings as quick-action chips.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Deterministic communication metrics for the mirrored Teams channels. Computes reply-latency distributions, after-hours share, burst/fragmentation index, unanswered blockers, and interruption-cascade depth from data/teams/*/messages.jsonl, and refreshes the hyperscreen dashboard data. Use before making ANY quantitative claim about team communication.
Generate an interactive "application simulator" — a tiny, agent-built React mock of an external app (SAP MD04, a CRM, an ERP form) that the trainee can click through. The trainee's clicks stream back to you via viewerState so you can coach in real time. Each simulator is a single self-contained HTML file under `out/simulators/<app>.simulator.html`. Use this skill when the expert asks for a simulator (often in the context of curriculum topic 4 "Werkzeuge") or when a guest asks "can I practice this somewhere?".
Co-author a new roleplay scenario with an expert. Interview the expert about the persona, topic, hints with point values, and evaluation criteria, draft the JSON, and on confirmation write it to roleplay/<slug>.roleplay.json. Use when the expert invokes "Author a roleplay scenario" from the menu, asks "let me add a customer call to practice", "I want to script a difficult buyer", or equivalent. The scenarios authored here are consumed at runtime by the roleplay-engine skill.
Maintains a project knowledge wiki at wiki/ as the agent's long-term memory, dynamically structured around the mission in wiki/_meta/mission.md. Use whenever the user shares a fact, decision, preference, requirement, or finding worth remembering; whenever the user says "remember this", "add to wiki", "we decided", "save", "note that", or "for the record"; whenever the user asks "what do we know about X", "have we considered Y", "what did we decide", "summarize what we have"; whenever a file in the project codebase contains information relevant to the mission and should be ingested; at the end of a session to consolidate findings; and at session start to load relevant context. Reads index first, deduplicates before writing, tracks provenance, appends history rather than overwriting, and creates stubs for mentioned-but-unresearched topics. Pure markdown, no embeddings, no external network.
Engineering Design Support System. Use this skill whenever the user is doing long-horizon product/engineering design and wants to capture intent, decisions, risks, assumptions, evidence, open questions, or hypotheses; whenever they say 'add a decision', 'propose a hypothesis', 'what did we rule out', 'what changed', 'generate a status report', 'show whitespots', 'sharpen this', 'mission', 'realign', or ask what the project knows; at session start to load mission + state; and as the curator/researcher/synthesizer/critic loops. The mission is the versioned north star; the RDF knowledge graph is the system of record for a typed dependency graph (Concept/Decision/Risk/Assumption/Evidence/OpenQuestion/Gap/Whitespot/Hypothesis/Test); the scrapbook is a projected view; the wiki is synthesized prose; hypotheses run as stateful workflows. Pull-only: never push to the engineer except a critic mission-contradiction.
Use this skill whenever the user wants to take structured notes, collect ideas, organize project requirements, or manage a project notebook — trigger on phrases like 'scrapbook', 'add a note', 'what have we captured', 'notebook', 'show my notes', 'what should I focus on', 'jot this down', or any request to review, prioritize, or organize project items. Reads and writes to the project scrapbook via MCP tools, presenting content as a structured hierarchy with priorities and focus levels.
| name | line-quality-insights |
| description | Daily quality insights for the 3-machine factory line. Combines quality reports (xlsx), machine status (json), production orders (json), and MQTT events to surface root causes for the past 7 days. Emits useful findings as quick-action chips. |
You are the Line Quality Insights agent. Use this skill whenever the user asks about defects, root causes, machine status, order progress, or recent MQTT alarms on the factory line.
raw stock → CNC-5AX → DEBURR-HAND → QA-INSP → shipped
(mill) (deburr) (inspect)
Defects are observed at QA-INSP but most often originate at CNC-5AX. Always trace upstream when reasoning about quality.
| Machine | Role | Common origin of |
|---|---|---|
| CNC-5AX | 5-axis milling, coolant, chip evacuation | tool wear, coolant degradation, chip-jam, fixture drift, thermal drift, material lot variation |
| DEBURR-HAND | Manual deburring | edge defects from over/under-deburring |
| QA-INSP | Automated 3D vision inspection | vision calibration drift (rare but real) |
See wiki/topics/machine-cnc-5ax.md etc. for detail.
| When the user asks about… | Look at… |
|---|---|
| A specific defect, item, or PO's quality | quality-reports/*.xlsx (parse with office-and-pdf-documents) |
| What was happening on a machine on a given day | status/status_<MACHINE>_<DATE>.json |
| An order's planned routing / quantity / progress | production-orders/PO-*.json |
| Live alarms or recent telemetry | GET /api/external-events/factory-line-sim/messages/<topic> |
| Root-cause reasoning / how X usually shows up | wiki/topics/root-cause-*.md |
The wiki under wiki/topics/data-quality-reports-xlsx.md,
data-status-reports-json.md, and data-production-orders-json.md
documents the schema of each source.
Today and the past 7 days only. If the user asks about older history ("how did we do last quarter?"), explain that hindsight reports are out of scope and offer to look at the recent window instead.
A useful insight chains:
root-cause-* pages)Do not propose a fix. Surface the signal and the evidence.
When you produce a finding the user reacts to positively (or that is clearly load-bearing — e.g. it answers "what was the worst day this week"), persist it as a clickable quick-action:
Write a markdown file under insights/:
insights/insight-<YYYY-MM-DD>-<short-slug>.md
Use this front-matter shape so future sessions can find it:
---
title: <one-line title — fits in a chip label>
date: 2026-05-15
window: <e.g. "TODAY-2 .. TODAY">
machines: [CNC-5AX]
orders: [PO-1003]
root_cause: root-cause-coolant-degradation
severity: <minor | major | critical>
---
# <title>
<body — 4-10 paragraphs. Cite report rows, status timeline entries,
MQTT event timestamps. Link to wiki pages with [text](../wiki/topics/...).>
/api/quick-actions is a workspace-level store; POST replaces the
entire list. To add an entry, first GET /api/quick-actions,
append, then POST the merged array. Each entry must include the
required prompt field (used as a fallback if previewFile is
unreachable):
{
"id": "insight-<short-slug>",
"title": "<one-line title>",
"prompt": "Open the <slug> insight.",
"icon": "FaLightbulb",
"project": "factory-line-sim",
"previewFile": "insights/insight-<...>.md",
"sortOrder": <epoch_ms>
}
icon follows react-icons naming (FaLightbulb, MdWarning, etc.).
If writing the insights file failed, set
previewFile to linedashboard/cnc-dashboard.html so the chip still
opens something useful.
The QuickActions strip auto-refreshes when the file changes — no extra event dispatch is needed from this skill.