This workflow is for anyone who runs n8n workflows and wants to know immediately when something breaks, without digging through execution logs. Especially useful for non-technical users and freelancers or agencies managing workflows for clients.
How it works
When a workflow fails, this system activates automatically. Claude AI analyzes the error using the technical details and the business context you provide, then generates a diagnosis with the direct cause, a step-by-step fix and prevention advice. A second AI agent validates the confidence score. Based on that score you receive a full diagnosis (≥80%), a diagnosis with a reservation (50–79%), or a manual review notification (<50%). All diagnoses are stored in your Supabase database.
Nice share. One fast way to make this even more useful is to add a quick triage split before deeper checks:
classify the error first (auth, schema/field mismatch, rate-limit, timeout, downstream validation),
run one minimal check per class so users can rule out common causes in under 5 minutes,
log the exact failing node + expected vs actual field for the first failing path.
If someone has one concrete broken path, I can help scope a 100 USD Automation Fix Sprint: one workflow path, diagnosis, smallest safe fix if possible, verification checklist, and short handoff note.
The two-agent confidence validation pattern is a smart design choice - using a second AI to sanity-check the diagnosis before sending it out prevents noisy or low-confidence alerts from cluttering inboxes. Using Supabase as the persistent store also means you build a searchable error history over time, which is genuinely useful for spotting recurring failure patterns across workflows.
Recently built an automated error diagnosis workflow for faster debugging and troubleshooting.