PURPOSE

Designs and outputs n8n workflow JSON with robust triggers, idempotency, error handling, logging, retries, and human-in-the-loop review queues.

WHEN TO USE

• TRIGGERS:

- Build an n8n workflow that runs every Monday and emails the compliance summary. - Add error handling and retries to this workflow, plus a review queue for failures. - Create a webhook workflow that logs every run and writes a status row to a tracker. - Make this n8n flow idempotent so it does not duplicate records when it reruns. - Instrument this workflow with audit logs and a human approval step.

• DO NOT USE WHEN…

- You need code-only automation without n8n (use a scripting/CI skill). - You need to bypass security controls or hide audit trails. - You need to purchase or recommend prohibited items/services.

INPUTS

• REQUIRED:

- Workflow intent: trigger type + schedule/timezone + success criteria. - Targets: where to write results (email/Drive/Sheet/DB) and required fields.

• OPTIONAL:

- Existing n8n workflow JSON to modify. - Sample payloads / example records. - Definition of dedup keys (what makes a record unique).

• EXAMPLES:

- Cron: Monday 08:00 Europe/London; send summary email + Drive upload - Webhook: receive JSON; route to folders

OUTPUTS

• Default (read-only): a workflow design spec (nodes, data contracts, failure modes).
• If explicitly requested: workflow.json (n8n importable JSON) + runbook.md (from template).

Success = workflow is idempotent, logs every run, retries safely, and routes failures to a review queue.

WORKFLOW

• Clarify trigger:

- Cron/webhook/manual; schedule/timezone; concurrency expectations.

• Define data contract:

- input schema, required fields, and validation rules.

• Design idempotency:

- choose dedup key(s) and storage (DB/Sheet) to prevent duplicates on retries.

• Add observability:

- generate run_id, log start/end, store status row and error details.

• Implement error handling:

- per-node error branches, retry with backoff, and final failure notification.

• Add human-in-the-loop (HITL) review queue:

- write failed items to a queue (Sheet/DB) and require approval to reprocess.

• “No silent failure” gates:

- if counts/thresholds fail, stop workflow and alert.

• Output:

- If asked for JSON: produce importable n8n workflow JSON + runbook.

• STOP AND ASK THE USER if:

- destination systems are unknown, - no dedup key exists, - credential strategy (env vars) is not specified, - the workflow needs privileged access not yet approved.

OUTPUT FORMAT

If outputting n8n workflow JSON, conform to:

{
  "name": "",
  "nodes": [ { "name": "Trigger", "type": "n8n-nodes-base.cron", "parameters": {}, "position": [0,0] } ],
  "connections": {},
  "settings": {},
  "active": false
}

Also output runbook.md using assets/runbook-template.md.

SAFETY & EDGE CASES

• Read-only by default; only emit workflow JSON when explicitly requested.
• Do not include secrets in JSON; reference env vars/credential names only.
• Include audit logging + failure notifications; avoid workflows that can silently drop data.
• Prefer least privilege: call only required APIs and minimize scopes.

EXAMPLES

• Input: “Cron every Monday, email compliance summary, retry failures.”

Output: Node map + workflow.json with Cron → Fetch → Aggregate → Email, plus error branches to review queue.

• Input: “Webhook that logs runs and writes status row.”

Output: Webhook → Validate → Process → Append status row; on error → log + notify + queue.