agentsclimarketplace

Leadup n8n workflow builder

Skill leadupofficial/leadup-claude-skills/leadup-n8n-workflow-builder

Plan an n8n workflow for a LeadUp business or client — connecting webhooks, APIs, CRMs, WhatsApp BSPs, Google Sheets, email, payment events, and AI calls into a clear graph with triggers, nodes, data mapping, credential placeholders, error handling, manual approval points, a test plan, and a text-form workflow diagram. Use when the user says "n8n workflow", "automation workflow", "build workflow", "webhook automation", "connect APIs", or "workflow builder".From its SKILL.md

Install
npx -y skills add leadupofficial/leadup-claude-skills --skill leadup-n8n-workflow-builder

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.5 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

LeadUp n8n Workflow Builder

Purpose

Design an n8n workflow LeadUp can paste into the editor: trigger, nodes, data mapping, credentials (as placeholders), error handling, manual approval points, and a test plan. Output works for cloud n8n or self-hosted on leadup-server.

When to use

Use when the user wants an n8n graph designed. Do not trigger for the WhatsApp template / opt-in design only (use leadup-whatsapp-automation-planner), AI feature design (use leadup-ai-feature-planner), or the broader funnel (use leadup-lead-funnel-builder).

Trigger phrases: "n8n workflow", "automation workflow", "build workflow", "webhook automation", "connect APIs", "workflow builder", "n8n graph", "automate this".

Inputs needed

  • Business outcome the workflow drives (lead capture, reminder send, CRM sync, invoice creation, AI-summary on event, etc.).
  • Trigger source (webhook / cron / inbound email / app event / manual).
  • Systems involved (CRM, BSP, Razorpay, Google Sheets, Stripe, Slack, Anthropic API, OpenAI, n8n itself).
  • Data shape (what comes in, what goes out, what must be validated).
  • Volume / rate expectations.
  • Failure tolerance (must-not-miss / can-retry / fire-and-forget).
  • Approval requirement (human in the loop?).

Ask at most 2 clarifying questions if the trigger or systems are unclear.

Tools/resources to use

  • references/n8n-workflow-framework.md — node patterns, error handling, credential rules.
  • assets/n8n-workflow.template.md — output shape (workflow as text graph).
  • leadup-api-research-builder — for each external API.
  • leadup-whatsapp-automation-planner — for the WhatsApp side.
  • leadup-ai-feature-planner — if an LLM call lives inside the workflow.
  • leadup-pii-risk-reviewer — if customer data passes through.
  • leadup-security-review — for payments / sensitive scopes.
  • leadup-status-updater — to record the new automation in project memory.

Step-by-step workflow

  1. Restate the workflow outcome in one line.
  2. Pick the trigger (webhook URL pattern / cron / event / inbound).
  3. Map data shape end-to-end: input → transforms → outputs.
  4. Plan nodes in order; one job per node.
  5. Credentials: list each required credential with a placeholder ({{n8n_creds.crm_token}}); never paste real values.
  6. Error handling: per-node retries, on-error branch (Slack alert, dead-letter Google Sheet), idempotency keys.
  7. Manual approval point: add a wait-for-webhook / approval step for sensitive actions (refund, mass email, destructive).
  8. Test plan: 3–5 input fixtures + expected outputs.
  9. Workflow diagram (text): a clear textual flow the team can recreate in the n8n editor without ambiguity.

Required output format

One Markdown plan with these sections, in this order:

  1. Brief restate — outcome, trigger, systems.
  2. Trigger — type, URL pattern (if webhook), schedule (if cron), sample payload.
  3. Data shape — input JSON schema, key transforms, output schema.
  4. Nodes (ordered) — list with name · type · purpose · input · output.
  5. Credentials — list with placeholder name and where the secret lives (vault / env / n8n credentials store).
  6. Error handling — per-node + workflow-level (alerts, dead-letter, idempotency).
  7. Manual approval points — where + who + via what channel.
  8. Test plan — fixtures with expected outputs.
  9. Workflow diagram (text) — ASCII/Markdown graph the team can recreate.
  10. Hand-offs — to other LeadUp skills.

Safety rules

  • Do not paste real API keys, tokens, webhooks, or customer data. Use placeholders.
  • Do not log full PII in n8n executions. Mask phone / email / payment fields; truncate IDs.
  • For payment-changing nodes (Razorpay refund, Stripe charge), add a manual approval step or restrict to a verified caller.
  • Webhook endpoints must verify signature where the source supports it (Razorpay HMAC, Stripe signature, WhatsApp BSP signature, GitHub X-Hub-Signature).
  • Idempotency: dedupe on a stable key (event id, message id) before side-effect nodes.
  • Retries with backoff for transient errors; dead-letter for permanent.
  • For India: respect IT Rules / DPDP scope on customer data; document consent gate where relevant.
  • Defer LLM API specifics to leadup-api-research-builder and leadup-ai-feature-planner.

Common mistakes

  • One node doing 5 jobs — hard to debug.
  • No idempotency → duplicate side effects on retries.
  • Real secrets pasted into "HTTP Request" node URL.
  • No error branch → silent failures.
  • No retry policy → flaky third-party kills the workflow.
  • Manual triggers used in prod for time-sensitive flows.
  • No test fixtures → "it worked once" is the only proof.
  • Workflows that hit rate limits without backoff.

Troubleshooting

  • Self-hosted n8n on leadup-server: confirm container resources, webhook public URL via Coolify, queue mode for high volume.
  • Cloud n8n: confirm execution data retention + plan limits.
  • High volume: switch to queue mode, batch nodes, or off-load to a dedicated worker.
  • AI inside workflow: cap token use per execution; cache by input hash; route through leadup-ai-feature-planner for cost ledger.
  • Cross-system data shape mismatch: add a "Set" / "Function" node early to normalize; document the canonical shape.
  • Manual approval needed sometimes only: use a router (if condition → approval; else → continue).

Test prompts

Should trigger (5)

  1. "Build an n8n workflow that captures lead form → CRM → WhatsApp follow-up."
  2. "Automate Razorpay payment webhook → invoice → email."
  3. "n8n graph: Google Sheet new row → AI summary → Slack."
  4. "Workflow to sync new bookings to Google Sheets every hour."
  5. "Connect our CRM to WhatsApp BSP via n8n."

Should NOT trigger (3)

  1. "Design the WhatsApp template + opt-in." (→ leadup-whatsapp-automation-planner)
  2. "Plan the AI feature itself." (→ leadup-ai-feature-planner)
  3. "Research the Razorpay webhook docs." (→ leadup-api-research-builder)

Functional test cases (2)

  1. Given "lead form on website → CRM upsert → WhatsApp Utility template → Google Sheet log", return an ordered node list, credential placeholders, signature-verified webhook trigger, idempotency key on submission id, retry + dead-letter on downstream failures, fixtures with expected outputs, and a text workflow diagram.
  2. Given "Razorpay payment webhook → refund a payment under conditions", return a plan with HMAC signature verification, manual-approval node for any refund, audit-log node, and a hand-off to leadup-security-review and leadup-pii-risk-reviewer.

Success criteria

  • All 10 sections present in order.
  • Nodes are atomic (one job each).
  • All credentials are placeholders.
  • Webhook signature verification specified.
  • Idempotency key chosen.
  • Error handling per node + workflow-level.
  • Manual approval used for sensitive actions.
  • Test fixtures with expected outputs.
  • Text workflow diagram unambiguous.
  • Hand-offs explicit.

What ships with it: 2 files

8.9 KB alongside SKILL.md

references/

Keep looking

Skills are one crate of 326,367. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.