Automation Workflow Design

Automation Workflow Design

Before you start

Gather the following from the user:

1. What is the trigger? (Webhook, schedule, database event, form submission, file upload)
2. What is the desired outcome? (Record created, notification sent, file transformed, data synced)
3. Which platform? (n8n, Make, Zapier, Power Automate, or platform-agnostic)
4. What systems are involved? (APIs, databases, SaaS tools, file storage)
5. What is the expected volume? (Runs per hour/day, payload sizes)
6. What happens on failure? (Retry, alert, fallback, manual queue)

If the user says "automate this process," push back: "Which trigger starts it, what systems are involved, and what should happen when a step fails?"

Workflow mapping procedure

Step 1: Define the trigger

Identify the event that starts the workflow. Document:

• Trigger type: Webhook, polling, cron schedule, event subscription
• Payload shape: Fields available from the trigger (provide a sample JSON)
• Frequency: Expected invocations per time window
• Idempotency: Whether duplicate triggers are possible and how to handle them

Step 2: Map the action chain

For each step after the trigger, document in sequence:

Step [N]: [Action name]
  Platform node: [Specific node/module name for the target platform]
  Input: [Which fields from previous steps]
  Transform: [Any data mapping or formatting]
  Output: [Fields produced for downstream steps]
  Error behavior: [Retry N times / skip / halt workflow]

Group related steps into logical stages: Intake, Processing, Output, Notification.

Step 3: Add conditional branches

For each decision point:

• State the condition as a boolean expression using available fields
• Document the true-path and false-path step sequences
• Identify whether branches rejoin or terminate independently

Step 4: Design error handling

For every step that calls an external service:

• Retry strategy: Count, backoff interval, max wait
• Timeout: Maximum seconds before treating as failure
• Dead letter handling: Where failed items go (error queue, log table, notification)
• Circuit breaker: If failure rate exceeds threshold, pause workflow and alert

Step 5: Draw the workflow diagram

Produce a text-based diagram showing the flow:

[Trigger: New form submission]
  |
  v
[Validate payload] --invalid--> [Log error + notify]
  |valid
  v
[Enrich: Fetch customer from CRM]
  |
  v
[Branch: Is enterprise?]
  |yes                |no
  v                   v
[Create Jira ticket]  [Add to Mailchimp list]
  |                   |
  v                   v
[Notify #sales]       [Send welcome email]

Include error paths as dashed or labeled branches.

Platform-specific notes

n8n: Use Error Trigger node for global error handling. Reference nodes by exact name (HTTP Request, IF, Switch, Merge). Use expressions with {{ $json.field }} syntax.

Make (Integromat): Use error handler routes (Break, Resume, Ignore, Rollback). Reference modules by app name + action. Note the 15-minute default scenario timeout.

Zapier: Use Paths for conditional logic. Note that error handling requires Formatter + Filter workarounds or custom code steps. Multi-step zaps have a 30-second per-step timeout.

Power Automate: Use Scope actions for try-catch patterns. Configure run-after settings (succeeded, failed, skipped, timed out) on each action. Use @outputs('step')['statusCode'] for error inspection.

Workflow document template

# Workflow: [Name]

## Overview
- **Trigger**: [Event description]
- **Platform**: [n8n / Make / Zapier / Power Automate]
- **Schedule**: [If applicable]
- **Owner**: [Team or person]

## Steps
| # | Action | Node/Module | Input | Output | Error Handling |
|---|--------|-------------|-------|--------|----------------|
| 1 | ...    | ...         | ...   | ...    | ...            |

## Branches
| Condition | True path | False path |
|-----------|-----------|------------|

## Error handling
- Global error handler: [Description]
- Dead letter destination: [Queue / table / channel]
- Alert channel: [Slack / email / PagerDuty]

## Volume & limits
- Expected runs: [N per hour/day]
- Rate limits: [Per-API limits to respect]
- Payload size limits: [If applicable]

Quality checklist

Before delivering the workflow design, verify:

• [ ] Every step has explicit input fields, output fields, and error behavior
• [ ] Conditional branches document both paths and specify whether they rejoin
• [ ] Error handling covers retries, timeouts, dead letters, and alerting
• [ ] The workflow diagram matches the step table exactly
• [ ] Platform-specific node names and syntax are correct for the target platform
• [ ] Idempotency is addressed for the trigger and any create/update operations
• [ ] Volume estimates and rate limits are documented

Common mistakes

• No error handling on HTTP steps. Every external call can fail. Skipping retry and timeout config means silent data loss in production.
• Ignoring idempotency. Webhooks can fire twice. Without deduplication (check by ID before create), you get duplicate records.
• Hardcoding credentials in workflow steps. Use the platform's credential store or environment variables. Never embed API keys in node configurations.
• Linear-only thinking. Real workflows branch. If you have zero conditional nodes, you probably have not considered validation failures or edge cases.
• Missing the "what if the whole workflow fails" path. Individual step retries are not enough. Design a global error handler that captures the workflow run ID, failed step, and payload for manual replay.
• Not documenting rate limits. A workflow that runs fine at 10/hour breaks at 1000/hour when a batch import triggers it. State the ceiling.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: The-AI-Directory-Company
• Source: The-AI-Directory-Company/agents-and-skills
• License: MIT
• Homepage: https://ai-directory.company

Install and usage instructions live in the source repository linked above.