N8n pro

n8n-pro Skill

Purpose

Operate, harden, monitor, and back up the Weblyfe n8n fleet at app.n8n.weblyfe.nl. Codifies the patterns and tools built 2026-05-02 so future sessions don't have to re-derive them.

Quick reference

Task	Tool / Command
Daily JSON backup of every workflow to git	tools/n8n-backup.py (launchd com.weblyfe.appie-1-cron.n8n-backup at 02:15)
Hourly health monitor + Telegram alerts	tools/n8n-health.py (launchd com.weblyfe.appie-1-cron.n8n-health every 3600s)
Apply retry defaults to every active workflow	tools/n8n-harden.py [--dry-run] [--workflow ID]
Inspect workflow JSON via API	curl -H "X-N8N-API-KEY: $KEY" https://app.n8n.weblyfe.nl/api/v1/workflows/<id>
List + count active workflows	curl ...?active=true&limit=100
Activate / deactivate a workflow	POST /api/v1/workflows/<id>/activate (or /deactivate)
Update a workflow	PUT /api/v1/workflows/<id> with body {name, nodes, connections, settings} only — anything else triggers request/body must NOT have additional properties

Configuration

API base + key live in ~/.openclaw/openclaw.json under env.vars:

N8N_BASE_URL  # "app.n8n.weblyfe.nl"
N8N_API_KEY   # Bearer-style token, sent as X-N8N-API-KEY header

DigitalOcean droplet: n8n-weblyfe (id 545108451, region ams3, size s-1vcpu-1gb, public IP 142.93.232.231). DO API token in ~/.openclaw/openclaw.json as DIGITALOCEAN_ACCESS_TOKEN. No SSH key registered on DO account, so direct SSH from Appie-1 to the droplet is blocked — see Notion task 354c3321-de60-8128-b1b1-df81e0ddc0ed for the manual key-add procedure.

Production-ready workflow checklist

Before declaring a workflow "foolproof," verify all of the following:

1. Error handler is wired. settings.errorWorkflow points to the GLOBAL ERROR HANDLER 54kH3T3oXWELrCDuFKMN5. The handler auto-attaches itself to any active workflow without one (Schedule Trigger at 1am daily), so this is mostly automatic — but verify on PR.
2. Retries on every API node. retryOnFail: true, maxTries: 3, waitBetweenTries: 2000 minimum. Use tools/n8n-harden.py --workflow <id> to apply.
3. Graceful degradation on non-critical leaf nodes. Sheet appends and label-adds get onError: continueRegularOutput or continueErrorOutput so a downstream service hiccup doesn't block the upstream archive.
4. Idempotent file naming. Drive uploads use {sender_domain}_{email_date}_{thread_id_short}.pdf style — not {$now} — so retries don't create duplicates.
5. Subject filter is specific. A trigger with only has:attachment will hit every random PDF (contracts, screenshots, your own outbound mail). Add explicit subject patterns.
6. Skip rules for known false positives. Stripe TECHWIZ LLC = income, skip. Outbound mail from [email protected] = own content, skip. Embed in a Code node early in the flow.
7. Multi-attachment handling. If a single email might carry multiple PDFs (Tally, Midjourney), the Code node must emit one item per PDF — n8n's Object.keys(...).find(...) only takes the first.
8. Folder mapping covers backfill + future. The Drive folder map should include 2024-2026 and a fallback that auto-discovers via Drive search.
9. Single-account vs multi-account. Each Gmail Trigger is bound to one credential. Multi-account coverage requires either multiple Trigger nodes or a parallel pipeline (Python tools/invoice-pipeline.py covers all 3 accounts as backup).

Common API gotchas

• PUT body schema is strict. Send {name, nodes, connections, settings} only. Any of id, active, createdAt, updatedAt, versionId, tags, triggerCount, meta, staticData, isArchived, shared, homeProject, scopes, sharedWithProjects, pinData, usedCredentials causes 400 request/body must NOT have additional properties.
• PUT validates credentials. If a node references googleContactsOAuth2Api (or any other credential type) and that credential no longer exists in the workspace, PUT returns 400 Cannot publish workflow: N nodes have configuration issues. Fix in the n8n UI before harden can succeed.
• Pagination. List endpoints page via nextCursor — always loop until cursor is empty.
• Activate vs PUT-with-active. Use the dedicated POST /workflows/<id>/activate endpoint; setting active: true in PUT body is rejected (it's a read-only field on PUT).

Workflow node-type cheatsheet

API-touching (always retry):

n8n-nodes-base.gmail / .gmailTrigger
n8n-nodes-base.googleDrive / .googleSheets / .googleCalendar / .googleContacts
n8n-nodes-base.airtable / .notion / .slack / .telegram
n8n-nodes-base.httpRequest / .webhook
n8n-nodes-base.stripe / .mailerLite / .brevo / .openAi
n8n-nodes-base.tally / .tidyCal / .calendly / .intercom
@n8n/n8n-nodes-langchain.*  (LLM nodes)
n8n-nodes-base.extractFromFile  (PDF parsing — IO-bound)

Local logic (skip retry):

n8n-nodes-base.set / .code / .function
n8n-nodes-base.if / .switch / .merge
n8n-nodes-base.splitInBatches / .wait / .noOp
n8n-nodes-base.scheduleTrigger / .cron
n8n-nodes-base.manualTrigger / .errorTrigger

tools/n8n-harden.py enforces this split via regex patterns.

Backup + rollback procedure

Backups live at ~/clawd/n8n-backups/<workflow_id>__<slug>.json. Daily commit by tools/n8n-backup.py.

To roll back a single workflow to its prior state:

# Find the previous version
git -C ~/clawd log -p -- n8n-backups/<id>__*.json | less

# Restore from a specific commit
git -C ~/clawd show <commit>:n8n-backups/<id>__<slug>.json > /tmp/restore.json

# PUT the trimmed body back via API
python3 -c "
import json
wf = json.load(open('/tmp/restore.json'))
keep = {k: wf[k] for k in ('name','nodes','connections','settings') if k in wf}
json.dump(keep, open('/tmp/put.json','w'))
"
curl -X PUT -H 'X-N8N-API-KEY: $N8N_API_KEY' -H 'Content-Type: application/json' \
  --data-binary @/tmp/put.json \
  https://app.n8n.weblyfe.nl/api/v1/workflows/<id>

Health monitoring

tools/n8n-health.py (hourly):

• Failure check: queries /executions?status=error&limit=50, filters to last 1h window, aggregates by workflow, alerts via Telegram MarkdownV2.
• Silent-death check: queries last 200 executions, finds workflows with scheduleTrigger/cron that have NO successful run in 24h, alerts.
• Cooldown: 30 min per alert key (state in ~/clawd/.cache/n8n-health-state.json) so flapping doesn't spam.

State of failing/silent workflows visible in ~/clawd/logs/n8n-health.jsonl. Common patterns:

• Whatsapp Agent | DRP failures — likely external Whatsapp API issues; check execution detail.
• GLOBAL ERROR HANDLER errors — usually a downstream issue that the handler couldn't auto-fix.

When to add a new workflow vs modify an existing one

• Add new when the trigger source is genuinely different (a new form, a new webhook URL, a new schedule). Naming convention: {Domain} | {What it does} (e.g., PeakSpring | Contact Form).
• Modify existing when the upstream/downstream stays the same and only the body logic changes.
• Clone-then-modify when a major refactor is needed (the Gmail Invoicing v2 example): keep the original active until v2 is verified, then deactivate the original. Old workflows are kept around as inactive (don't delete) so we have history.

Authoritative refs

• Doctrine: ~/clawd/knowledge/token-reduction-system/95-token-reduction.md (output discipline + tier routing — applies to LLM nodes inside n8n flows).
• This skill: ~/clawd/skills/n8n-pro/SKILL.md (you are here).
• Memory pointer: auto-memory reference_invoice_pipeline.md carries the fleet-wide ops summary.
• Backup history: ~/clawd/n8n-backups/ (47+ workflow JSONs, daily commits on master).
• Tools: tools/n8n-backup.py, tools/n8n-health.py, tools/n8n-harden.py, tools/invoice-pipeline.py.

Recent landmark events

• 2026-05-02 morning. Gmail Invoicing v1 (caS-m8n9KKb4tHBobvSgS) had a Q2 2026 folder ID typo (Uvxq vs real Uvx) plus 1x/day polling plus no archive-after-process. Patched in-place, then cloned to v2 (pgKuiPoPpmv9QKgl, "no AI"), then hardened with multi-PDF support + Stripe TECHWIZ skip + idempotent filename + tightened subject filter. v1 deactivated. Backup at commit aea94cc, post-state at 353d084 and 8b43952.
• 2026-05-02 same morning. GLOBAL ERROR HANDLER alert recipient changed from [email protected] solo to [email protected], [email protected] (both). Retry added to its Send a message node.
• 2026-05-02 same morning. First fleet-wide harden: 28 active workflows scanned, 20 patched (118 nodes received retry+wait defaults). 8 failed PUT validation due to missing credentials in pre-existing nodes — flagged for manual cleanup in n8n UI.