# Automation Prompt

> Turn a "when X happens, do Y" idea into a precise build prompt for an automation platform (n8n, Make, Zapier, and others). Trigger on "build an automation", "n8n prompt", "Make scenario", "Zapier prompt", "automate this workflow", "when X happens do Y", or any request to wire up a no-code or low-code workflow.

- Skill: `heath-gtm/automation-prompt` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add heath-gtm/automation-prompt`
- Raw SKILL.md: https://api.skillmd.com/api/skills/heath-gtm/automation-prompt/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: heath-gtm (https://skillmd.com/u/heath-gtm)
- Updated: 2026-08-19
- Page: https://skillmd.com/skills/heath-gtm/automation-prompt

---


# Automation Prompt

## What this does
Takes a workflow idea in plain words and returns a build prompt structured for an automation platform: the trigger, the steps in order, the data that moves between them, the branches, and the error handling. Automations break on the details. This pins them down before you build.

## What you'll need
An automation platform to build in (n8n, Make, Zapier, or similar). No connectors here; the prompt describes the workflow you then assemble. Bring the "when X, do Y."

## Customize this for yourself
| Set this | What it is | Example |
|---|---|---|
| Target platform | The tool you will build in | n8n, Make, Zapier |
| Your apps | The services in your stack the workflow will touch | Your CRM, your email, your sheet |
| Error policy | What should happen when a step fails | "Log it and notify me, do not retry silently" |

## The method
1. State the trigger precisely: what event starts this, and how often it can fire.
2. List the steps in order, each as "take this, do that, pass this along."
3. Name the data that moves between steps. The field that step 2 needs from step 1 is where automations break.
4. Add the branches: "if the record already exists, update it; if not, create it."
5. Add error handling: what happens when a step fails or returns nothing.
6. Add a test plan: the one input you will run through it to confirm it works end to end.

## Output (example)
Idea: "when a lead fills the form, add them to the CRM and Slack me."
Built prompt:
```
Platform: n8n.

Trigger: webhook, fires when the website form is submitted. Expect { name,
email, company, message }.

Steps:
1. Look up the email in the CRM (search by email).
2. Branch:
   - If found: update the record, append the message to notes.
   - If not found: create a new lead with the form fields.
3. Post to Slack #leads: "New lead: {name} at {company}. {message}"

Error handling: if the CRM step fails, post to Slack #alerts with the error and
the raw payload. Do not drop the lead silently.

Test: submit one form with a known email and one with a new email. Confirm one
updates and one creates.
```

## Example chat
**You:** When a lead fills out our form, I want them added to the CRM and pinged in Slack. Build me the automation prompt.

**Claude:** Here's the build prompt for n8n.
```
Trigger: webhook, fires when the website form is submitted.
Steps:
1. Look up the email in the CRM (search by email).
2. Branch:
   - If found: update the record, append the message to notes.
   - If not found: create a new lead with the form fields.
3. Post to Slack #leads: "New lead: {name} at {company}. {message}"

Error handling: if the CRM step fails, post to Slack #alerts with the error
and the raw payload. Do not drop the lead silently.
```
The detail that saves you a debugging session later: the error handling branch. Without it, a CRM hiccup just eats the lead and nobody notices until someone asks why a form fill never showed up.

**You:** What should I test before I trust it?

**Claude:** One form with a known email (should update) and one with a new email (should create). If both branches fire clean, you're live. One sharpener: tell me what happens if the same email submits twice in five minutes, that's the edge case most people forget to branch for.

## Go further
Hand-building one automation proves the shape. Here's the operator version.

- **Wire it straight into n8n or Make.** Paste the build prompt into the platform's AI workflow builder so the trigger, branches, and error handling assemble themselves instead of you clicking through nodes one by one.
- **Version the prompt, not just the workflow.** Keep the build prompt in a doc alongside the live automation, so when it breaks in six months you have the spec to rebuild from instead of reverse-engineering a tangle of nodes.
- **Chain automations off each other.** Once this one posts to Slack, add a second automation prompt that watches that channel and books a follow-up task in your CRM automatically.

The prompt is the blueprint. The platform just needs it spelled out once.

## Make it yours
Set your platform, your apps, and your error policy. Then describe any "when X, do Y" and get a build prompt with the data flow and the failure cases already mapped. Built by an operator. Customize it, break it, make it better.