Work public

n8n-self-healing

Self-healing n8n workflow. Learns from failures so the fourth identical error heals at zero LLM tokens. Claude, Slack, n8nac code-first.

  • n8n
  • self-healing
  • claude
  • slack
  • n8nac
  • automation
Repo
github.com/mj-deving/n8n-self-healing
Published
2026-05-26

The problem

n8n workflows fail. The same workflow tends to fail the same way. Most production teams trigger a Slack alert and have a human reroute the data. That works once. It does not scale to a fleet of workflows that fail dozens of times a day.

How it works

The workflow wraps its own execution in a recovery layer. When a step fails, the layer asks Claude to propose a fix, applies the fix, and records the failure signature plus the working recovery in a small recipe store. The next time the same failure signature appears, the layer looks up the recovery in the store and applies it directly. By the fourth identical error the recovery is free of LLM cost.

Failures the system cannot heal escalate to Slack with full context.

Stack

n8n workflow built code-first using the n8nac approach. Claude for the proposing layer. Slack for human escalation. The recipe store is the point.