n8n Automation Reliability Fixes

This repo is a sanitized case-study version of a production debugging job around an n8n reminder workflow that looked fixed in the editor but still behaved wrong in production.

What this repo shows

• how to debug the gap between editor state and real runtime state
• how to patch SQLite-backed workflow data carefully and reproducibly
• how to use independent signals instead of trusting the UI alone
• how to turn a one-off production fix into a repeatable operational workflow

The failure pattern

The workflow had several overlapping issues:

• edits were applied to the draft workflow, while the scheduler still used the active version snapshot
• a Schedule Trigger configuration looked valid in the UI but did not fire reliably in that specific setup
• callback state needed to be shared across workflows, not stored in per-workflow static data
• empty successful runs were hard to prove from execution history alone

What this repo contains

• small, sanitized Python scripts used to patch workflow JSON in SQLite-backed n8n
• examples of safer state-file usage for cross-workflow coordination
• a repeatable workflow for proving whether a trigger really fires

Screenshots

Reminder workflow

Subscription scanner workflow

Patch scripts

What this repo intentionally does not contain

• live n8n databases
• raw workflow exports
• webhook URLs, credential IDs, or instance-specific paths
• runtime logs and operational artifacts

Main lessons

1. In n8n 2.x, draft edits are not always the same thing as the active version the scheduler runs.
2. A visible UI change is not enough proof that the real runtime changed.
3. Cross-workflow state should live in a shared, mounted file or another explicit shared store.
4. A heartbeat file is a useful independent signal when execution history is incomplete.
5. Production fixes get safer when every patch script creates a backup first.