What it does: flowforge.run_failed fires when a run gives up, carrying the run, the flow, the index of the failed step and the error summary — so a flow of your own can page someone, open a ticket or post to a channel when another flow dies. Notes:

• It is free because the dead-letter queue is free: learning about your own outage is not something to charge for.
• It fires once per failed run. The event id is run_failed_ against a unique event hash, so one dead run cannot produce two alerts.
• A flow can never be started by its own failure: dispatch skips the failing flow’s own id, and a chain of failure-triggered flows is cut off at a depth of three (long enough for the legitimate “if the pager itself failed, send mail” pattern).
• The older one-argument flowforge/dead_letter action is kept for existing listeners.