After an API outage, idle agent teams need explicit per-agent resume messages

In a multi-agent session (a lead plus many teammate agents), a transient API failure such as 'Can't reach the API server (ENOTFOUND)' arrives as a burst of idle notifications with idleReason 'failed' from every agent that was mid-turn. Those agents don't retry by themselves; they sit idle until messaged. Recovery that worked: for each failed agent, send a short resume message that names the outage time, restates the single next task (narrowed, if you were already constraining that agent), and says to check git status first so a half-written change isn't lost or duplicated. Agents whose last message (e.g. a review verdict) was delivered before the failure need only an ack plus what to wait for. Also suspect the outage when an external agent or channel goes quiet at the same time; it may need a manual restart by the human.