Workflow Migrate — Automation Migration Tool

Why This Exists

Trigger

Invoked as: /workflow-migrate [workflow description or paste]

Process

Step 1: Parse the Workflow Input

From $ARGUMENTS:

• - If it's a description ("when a form is submitted, send email and update Airtable"): parse directly
• If it's N8N JSON: read and extract nodes/connections
• If it's Zapier steps: parse the trigger + action chain
• If it's Make (formerly Integromat) scenario: parse modules

Extract and document:

Triggers:

• - Webhook (POST endpoint)
• Cron/schedule (every X minutes/hours)
• Form submission
• Email received
• File drop (S3, Google Drive, local folder)
• DB row created/updated

Actions:

• - HTTP/API calls (list each endpoint, method, payload)
• Database reads/writes
• Email sends
• File operations
• Conditional logic (if/else branches)
• Loops over arrays
• Data transformations

Data flow:

• - Input payload fields used
• Intermediate computed values
• Output destinations

If the workflow description is too vague, ask 2-3 targeted questions before proceeding:

• - "What triggers it — webhook, schedule, or something else?"
• "Which API endpoints does it call and with what data?"
• "What counts as success? What should happen on failure?"

Step 2: Choose Language

Default to Python unless:

• - The workflow is heavily async/event-driven (prefer Node.js)
• Existing codebase is Node.js
• Kevin explicitly requests Node

Python stack: requests, schedule, logging, tenacity (retry), python-dotenv
Node.js stack: axios, node-cron, winston, async-retry, dotenv

Step 3: Write the Script

Generate a complete, runnable script. Required elements:

Python Template Structure:

Node.js Template Structure:

For each action in the workflow, write a dedicated function:

• - One function per logical action (e.g., fetch_leads(), send_notification(), update_database())
• Functions are composable and testable in isolation
• Every function logs what it's doing: start, result, any errors

Main orchestrator:
CODEBLOCK2

Self-healing patterns to include:

• - Alert on failure (Telegram message via Kevin's bot token if available, else log to file)
• Idempotency: skip already-processed records (use a state file or DB flag)
• Dead-letter queue: failed items saved to failed_[date].json for manual review
• Heartbeat: log "still alive" every N runs for scheduled workflows

Step 4: Generate .env.example

CODEBLOCK3

Step 5: Generate requirements.txt or package.json

Python:
CODEBLOCK4

Node.js:

{
  "dependencies": {
    "axios": "^1.6.0",
    "async-retry": "^1.3.3",
    "node-cron": "^3.0.3",
    "winston": "^3.11.0",
    "dotenv": "^16.3.1"
  }
}

Step 6: If the Workflow is Recurring — Generate a SKILL.md

If the workflow runs on a schedule or will be reused:

• - Generate a SKILL.md in the same output directory
• Name it after the workflow function
• Description: "Run [workflow name] — [what it does in 10 words]"
• Instructions: path to script, how to run it, what env vars are needed

Step 7: Save Outputs and Print Migration Summary

Output location: Ask Kevin where to save, default to INLINECODE17

Create:

• - workflow_[name].py (or .js)
• INLINECODE20
• INLINECODE21 (or package.json)
• INLINECODE23 (if recurring)
• INLINECODE24 (quick usage guide — 20 lines max)

Print migration summary:

Migration complete.

Original: [N8N/Zapier/Make] workflow — [X] nodes/steps
Output:   ./workflow_[name]/workflow_[name].py

What changed:
  - [X] N8N nodes → [Y] lines of Python
  - Added: retry with exponential backoff (3 attempts, 2s-30s)
  - Added: rotating log file (monthly)
  - Added: dry-run mode (--dry-run flag)
  - Added: failure alerts via [Telegram/log]
  - Removed: $XX/month SaaS subscription

To run:
  cd workflow_[name]/
  cp .env.example .env   # fill in your keys
  pip install -r requirements.txt
  python workflow_[name].py --dry-run   # test first
  python workflow_[name].py             # run for real

Error Handling

• - Ambiguous workflow: ask 2-3 targeted questions, don't guess at API endpoints
• Proprietary N8N nodes (e.g., OpenAI node): rewrite using the raw API — include API docs link in comments
• Very complex workflow (>15 nodes): break into multiple scripts with a coordinator, document the dependency order
• No credentials visible: generate .env.example with clear placeholders, add comments explaining where each key comes from
• Webhook trigger: generate a Flask/Express endpoint stub that receives the webhook and calls the workflow function

Notes

• - Always test with --dry-run before running for real
• State file for idempotency: ./state/processed_ids.json — create state/ dir if it doesn't exist
• For scheduled runs, prefer cron / schedule over external task runners (fewer dependencies)