automation
Automate a repetitive multi-step workflow across tools
Maps a manual, repetitive multi-step workflow into an automated one, with explicit failure handling and human checkpoints at the steps that need them.
Who should use this
Is this prompt for you?
- Founders or ops people manually repeating a multi-step process (data entry, reporting, notifications)
- Developers connecting multiple tools/APIs into one workflow (e.g. form -> CRM -> email -> spreadsheet)
- Teams that want to automate a process without losing visibility into what happened at each step
The prompt
Copy this prompt
Prompt
Help me automate the following manual workflow. Current manual process, step by step: [describe exactly what a human does today, in order, e.g. "1. New lead fills out a form. 2. I manually copy their info into the CRM. 3. I send them a welcome email. 4. I add them to a spreadsheet for weekly reporting."] Tools involved: [list the actual tools/APIs, e.g. "Typeform, HubSpot, Gmail, Google Sheets" or "our own database and SendGrid"] Trigger: [what should kick off the automation — a new form submission, a schedule, a webhook, a manual button] How this should run: [e.g. "a script triggered by a webhook", "a scheduled cron job", "a workflow tool like n8n/Zapier/Make" — specify if you have a preference] Design and build: 1. A step-by-step automation plan mapping each manual step to an automated action, and calling out any step that still needs a human decision (and shouldn't be fully automated) 2. The actual implementation (script, workflow config, or code) for the trigger and each step 3. Error handling for each step — what happens if step 2 fails after step 1 already succeeded? Does it retry, roll back, or notify a human? Be explicit, don't leave partial-failure behavior undefined 4. Idempotency — if the same trigger fires twice (e.g. a webhook retry), does the workflow avoid duplicating the action (double-emailing the lead, double-adding the row)? 5. Logging or a notification (e.g. Slack/email alert) when a step fails, so failures aren't silent 6. A rollback or manual-recovery plan for when the automation gets a run into a bad partial state List any step in the current manual process that should stay manual because it requires human judgment, and explain why automating it would be risky.
Works well with Claude Code, OpenAI Codex and Cursor.
How to use it
Getting the best result
- 1Write out your actual current manual steps in order — the automation should map to reality, not an idealized version
- 2Decide upfront which steps involve real judgment calls that shouldn't be fully automated, and say so
- 3Test the automation with a duplicate/retry trigger deliberately to confirm idempotency before relying on it
- 4Set up the failure notification (Slack/email) before turning on the automation, not after the first silent failure
Expected result
What you should get back
- A step-by-step automation plan mapping manual actions to automated ones, with human-judgment steps flagged
- Working implementation code or workflow configuration for the trigger and each step
- Explicit partial-failure handling — retry, rollback, or human notification defined per step
- Idempotency handling so duplicate triggers don't cause duplicate actions
Tips
Get more out of this prompt
- Always test what happens when step 2 of 4 fails — partial failure is the most common source of automation bugs
- Ask specifically about idempotency if any step sends an email, charges money, or creates a record — duplicates are expensive to clean up
- Keep at least one step manual if it involves real judgment (e.g. approving a large refund) rather than automating everything
Common mistakes
What to watch out for
- Automating the whole process without deciding what happens on partial failure, leading to silent data inconsistency
- Not testing for duplicate triggers (webhook retries are common) and ending up with duplicate emails or records
- Automating a judgment-heavy step that actually needed a human, removing an important check
Related prompts
Keep going
ai-agent
Design a scoped, tool-using AI agent for a specific task
Guides the design of a narrow, tool-using AI agent for one real task, including its system prompt, tool access boundaries, and failure/escalation behavior.
Copy promptapiIntegrate a third-party API with proper error handling and retries
Gets an AI to build a production-grade API client — authentication, retries, rate limiting, and typed responses — instead of a bare fetch call.
Copy promptpythonWrite a robust Python automation script with error handling
Produces a Python automation script that handles real-world edge cases — bad input, network failures, partial completion — instead of a fragile happy-path-only script.
Copy promptReady to put this prompt to work?
Pair it with an AI coding tool and a real project, then ship something this week.