Catching n8n workflows that run green but don't do their job

The failures that hurt me most in n8n weren’t red runs. They were green ones: the search came back empty, a login expired, a filter dropped every item, and the execution still said “Success”. An error workflow never fires for those.

So I built a watchdog workflow that looks at all published workflows through the n8n API every 15 minutes. How it works:

  1. Lists every published workflow via the n8n API, so there’s nothing to set up per workflow.
  2. Failed: any errored run gets an alert with the step that broke and the error message. Repeats are grouped into one alert with a count.
  3. Stopped running: for workflows with a Schedule Trigger, it checks whether the last run is overdue.
  4. Not doing its job: it checks whether successful runs actually reach the workflow’s last step. After about six good runs it learns each workflow’s normal rhythm (or you tag one, e.g. watch-24h).
  5. Stuck: a run still going after an hour (you can change that).
  6. A morning report on Telegram. If the report doesn’t arrive, n8n (or the Telegram link) may be down.

Failures are caught within 15 minutes. The others alert after a threshold: a stuck run after 1 hour, a stopped schedule after about 2 missed runs, a silent workflow once it’s quiet longer than usual (2 hours minimum).

It only reads; it never changes or stops anything, and it uses no AI or paid services. Tested on self-hosted n8n 2.41.5.

The idea you can steal even without the template: “reached the last node” is a much better health signal than “status = success”.

If you want the finished workflow plus a setup guide, it’s here: Workflow Watchdog: Telegram alerts when your n8n workflows fail, stop, or silently break (n8n template) (that link takes $10 off for the first 10 buyers)

Happy to answer questions about how the “not doing its job” check works.

silent green executions are far more dangerous than hard errors because monitoring dashboards assume everything succeeded normally.

adding an assertion node right before your termination step that validates non-empty record counts catches dropped payloads immediately.