A self-healing automation fleet

Problem

About 170 n8n workflows ran the company's operations: onboarding, recaps, billing alerts, community access, reporting. Only a handful had any error handling. When one failed it failed silently, and the team found out days later from a student or a missed report.

What I built

• One error handler for the whole fleet. Any failure posts a plain-language alert to a Slack channel and opens a round-robin ticket in the team's queue, with per-workflow de-duplication so a flapping job does not spam.
• Wired every active workflow to it through the platform's settings API, then set a team convention: every new workflow ships with the alarm, a time-saved estimate, and API visibility on.
• A nightly reconciliation job that catches the one failure class an error handler cannot see: a run that hangs instead of erroring. It re-derives the missing rows and repairs them.
• Encoding safety rules after a bulk edit corrupted non-ASCII text in dozens of workflows: every bulk change now goes through the settings API, never a full re-save, and a scan for corruption runs before anything is called done.

Result

Silent failures went to zero. Failures now surface within minutes with a ticket already assigned. The team stopped discovering outages from customers.

What it shows

Fleet-level thinking, not one workflow at a time. The alarm, the convention, and the scan are what make 170 workflows maintainable by two people.