Skip to main content

​
Creating a use case from scratch

The most common workflow: describe a business goal, get a production-ready use case. Your prompt: 
I need an automation that monitors our Gmail inbox for emails from clients,
extracts key information using AI, and logs everything in a Google Sheet.
If the email contains an attachment, save it to Google Drive.
What happens:
  1. The use-case-builder agent reads the platform documentation
  2. It designs an architecture with the right triggers, workflows, and integrations
  3. It presents a plan: “I’ll create 2 workflows — a Gmail service-event trigger for new emails, and a sub-workflow for AI extraction and file handling”
  4. After your approval, it creates the folder structure and delegates each workflow to n8n-workflow-builder
  5. It validates everything and shows you the complete use case
Tips for good prompts:
  • Describe the business goal, not the implementation (“monitor Gmail” not “use a Gmail trigger node”)
  • Mention the integrations you need (Gmail, Sheets, Drive, Slack, etc.)
  • Specify what should happen on success and on failure
  • Include any user-configurable settings (“the user should be able to choose which email label to monitor”)

​
Modifying an existing use case

When you need to add features, change triggers, or refactor. Your prompt: 
Add a Slack notification to the invoice-processor use case.
When an invoice is successfully processed, post a summary to a Slack channel.
What happens:
  1. The use-case-modifier reads the entire existing use case
  2. It identifies that a new workflow is needed (Slack notification) and that the main workflow needs to call it
  3. It presents a change plan: “I’ll add a new sub-workflow for Slack posting, update the main workflow’s success branch to call it, and add the Slack ORGCRED integration to config.ts”
  4. After approval, it makes targeted edits and delegates the new workflow to n8n-workflow-builder
  5. It validates to ensure nothing was broken

​
Building a single workflow

For when you need one workflow, not a full use case. Your prompt: 
Build an HTTP-triggered workflow that accepts a company name,
searches for it using Tavily, and returns a structured summary.
What happens:
  1. The n8n-workflow-builder reads the HTTP trigger guide and integration guides
  2. It creates a complete workflow JSON with the correct trigger, Codika Init, business logic, and Submit Result/Report Error pattern
  3. It handles all placeholder usage, credential configuration, and node positioning

​
Testing and debugging

After building (or when something breaks), use the tester. Your prompt: 
Test the invoice-processor use case and fix any issues.
What happens:
  1. The use-case-tester verifies and deploys the use case
  2. It triggers each HTTP workflow with test data constructed from inputSchema
  3. On success, it verifies outputs match outputSchema
  4. On failure, it fetches the execution trace, diagnoses the error, fixes the workflow, and redeploys
  5. It repeats up to 5 times until all workflows pass

​
The full build-test cycle

For maximum confidence, chain the agents together: 
Step 1: use-case-builder creates the use case
Step 2: use-case-tester deploys and tests it
Step 3: use-case-modifier fixes any remaining issues
Step 4: use-case-tester re-tests
You can do this in a single conversation: 
"Create a use case that [business goal], then test it end-to-end and fix any issues."

​
Writing effective prompts

​
Do

  • Be specific about integrations: “Use Gmail for email, Google Sheets for logging, and Slack for notifications”
  • Describe the data flow: “Extract the sender, subject, and key dates from each email”
  • Mention edge cases: “If the AI can’t parse the email, log it as unprocessed and skip”
  • Specify user settings: “The user should configure their Slack channel and email filter criteria at install time”

​
Don’t

  • Don’t specify implementation details: “Use a FLEXCRED_ANTHROPIC placeholder” — the agent knows this
  • Don’t prescribe trigger types: “Make it an HTTP POST webhook” — let the agent decide the best trigger
  • Don’t worry about Codika patterns: Init nodes, Submit Result, Report Error — the agent handles these automatically
  • Don’t specify node positioning or IDs: The agent follows standard conventions