Field notes
Local HITL sample
Another approval-gated path for calibration.
What Crash Labs saw
Gate FAIL · risk 100/100 · 2 outbound action(s) · coverage 100%.
"Very simple Human in the loop system email with AI e IMAP" should not go live yet: 3 critical issue(s) (score 100/100). Biggest risk: Nobody gets alerted when this workflow fails.
Why this breaks in production
critical
Retry survival: 2 side effect(s) still fire on delivery #2
Retry survival: 2 email · 2 side effect(s) survived run 2.
Fix → Add dedupe keys, idempotency, and rate limiting / a queue before outbound sends and Stripe writes.
critical
Simulation failed: Spam burst (10 triggers)
10 trigger fires produced 20 external action(s) (2+2+2+2+2+2+2+2+2+2 per run). 18 still fired after delivery #1. Ledger: 10 identical emails to info@n3witalia.com with subject "[Approval Required] Order question".
Fix → Add dedupe keys, idempotency, and rate limiting / a queue before outbound sends and Stripe writes.
critical
Fake-company ledger: 25× blast radius
25 identical emails to info@n3witalia.com with subject "[Approval Required] Order question".
Fix → Treat every outbound node as dangerous under retries. Require an idempotency key (order id / message id) before send/charge/write.
high
Nobody gets alerted when this workflow fails
You may think an email or invoice finished when it actually stopped halfway - and no one is paged.
Fix → In n8n Settings, set an Error Workflow (or add an Error Trigger flow) that Slack/emails your team with the failed run link.
Run this pattern on your workflow
Upload your n8n JSON at Crash Labs. Same gate, blast-radius ledger, and shareable report for clients.