n8n（n8n 自动化）

n8n — Dashboard-First Skill (Workflow Activity Dashboard)

Prime Directive (User Journey)

Your job is to get the agency user from Connected n8n → Live preview dashboard in minutes with low cognitive load:

1. Celebrate connection (no JSON)
2. Recommend Workflow Activity Dashboard by default
3. Auto-map fields silently (ask only when missing)
4. Show preview with real data
5. Offer optional "Turn a workflow into a product" next

For MVP, always start by selecting exactly one workflow that already has executions/events so the preview can be generated from real data immediately.

Never ask for UUIDs or show raw payloads unless user explicitly requests "raw".

Step 1 — "Connected!" Success Copy (use in chat + UI-friendly wording)

When n8n is connected and you have events/executions, summarize like:

🎉 n8n Connected

• ✅ Workflows indexed
• ✅ Execution activity detected
• ✅ Ready to build your Workflow Activity Dashboard

Detected signals (example phrasing):

• execution status (success/failure)
• execution duration (runtime)
• workflow identifier/name (grouping)
• timestamps (trend charts)
• error messages (for troubleshooting)

Primary CTA language: "Build Dashboard"

Step 2 — Default Template Choice

Default recommendation: Workflow Activity Dashboard (aka workflow monitoring / ops dashboard).

Choose this when the user says anything like:

• "monitor workflows", "activity dashboard", "failures", "reliability", "ops", "runs", "executions", "SLA", "debug", "what's breaking"

If the user asks for ROI: still start with Workflow Activity Dashboard, then add ROI widgets later (time saved, tasks automated estimates).

Tool Call Sequence (Agent MUST Follow This Order)

Required Tools for Dashboard Generation

Usage Rules

1. ALWAYS call analyzeSchema before any mapping tool
2. ALWAYS use generateMapping (id: generate-mapping) — NOT proposeMapping
3. NEVER assume field names — use only what analyzeSchema returns
4. NEVER proceed to preview if confidence < 0.75 — ask user to confirm missing fields first
5. If analyzeSchema returns fields from state JSONB (e.g., started_at, status, duration_ms), these are the normalized fields — prefer them over raw labels fields

Template ID Mapping

Step 3 — Auto-Mapping (what fields to look for)

Vocabulary

• workflow = automation definition
• execution = one run of a workflow
• node = a step inside a workflow
• trigger = how it starts (webhook/schedule/manual)
• status = outcome (success/error/waiting/running/crashed)

Minimal required signals for the dashboard

Map these with highest priority:

1. execution_status

• n8n: status
• normalize:

- success → completed - error / crashed → failed - waiting / running → running

1. timestamp

• prefer: stoppedAt for completed/failed executions
• fallback: startedAt if stoppedAt missing

1. duration_ms

• prefer: duration if present
• else: derive (stoppedAt - startedAt) in ms when both exist
• if still running: now - startedAt (optional)

Strongly recommended (for grouping + filtering)

• workflow_id: workflowId (or workflow name if that's what you have)
• trigger_type: mode (webhook/manual/trigger)
• execution_id: id

Step 4 — What to Show (widgets the user expects)

Use user-friendly labels:

• Total Executions (last 7d)
• Success Rate
• Failures (count)
• Avg Duration
• Executions Over Time (line chart)
• Recent Executions (table)
• Top Failing Workflows (bar list)
• Latest Errors (safe summaries)

If node-level data is available, offer optional drill-down: "slow node", "failed node" (don't assume it exists).

Step 5 — "Confirm Field Mapping" UX Rules

• 90% should be auto-matched.
• Only interrupt the user when required fields are missing or ambiguous.
• If missing, present 2–3 suggested fields (no manual typing first).

Ask in a UI-friendly way:

• "I can't find duration. I found: execution_time, runtime_ms, stopped_at-started_at. Which should we use?"

Pre-Preview Checklist (Validate Before Generating)

Before calling runGeneratePreviewWorkflow, verify:

• analyzeSchema returned fields from state JSONB (not just top-level columns)
• status field is mapped and has non-empty values
• started_at field is mapped (source: normalizer's state.started_at)
• ended_at field is mapped (source: normalizer's state.ended_at)
• duration_ms is available (computed by normalizer from timestamps)
• workflow_id is mapped for grouping
• Overall mapping confidence ≥ 0.75
• If confidence < 0.75: asked user about missing fields with 2-3 suggestions

If started_at or ended_at are missing but timestamp exists: → Offer: "I found timestamps but not execution start/end times. This means duration-based widgets will be empty. Continue anyway?"

Step 6 — Optional: Turn Into a Product (SaaS wrapper mode)

Only after dashboard preview is working, offer:

"Want to turn one of these workflows into a client-facing product (form → workflow → output), fully white-labeled?"

When user says yes:

• pick a single workflow
• define inputs (form fields)
• define output (success message, file, record created, webhook response)
• keep it branded and platform-hidden (do not mention n8n in client-facing copy)

Common Data Signals in n8n (don't overwhelm the user)

• Execution reliability: success/failure, retries
• Performance: duration, slow runs
• Volume: runs per day/hour
• Errors: message, failed node (if available)
• Triggers: webhook/manual/schedule

For deeper details, read:

• references/data-signals.md
• references/workflow-archetypes.md
• references/mapping-cheatsheet.md

What Not To Do

• Do not show raw JSON, schemas, UUIDs, or database details unless user asks.
• Do not assume node-level details exist (only use if present).
• Do not "invent" ROI numbers. If estimating time saved, label it as estimate and explain the assumption briefly.
• Do not break the dashboard-first flow by jumping to productization immediately unless user explicitly asks.