Replaying an event
swarm event replay <event-id> (the event.replay method) redelivers a persisted event to
its original subscribers, or a subset with --subscriber. It creates a new event that points
back at the original, queues fresh deliveries, and records the replay itself in the audit
trail. Only terminal originals (delivered, failed, or dead-lettered) are replayable.
Replaying for an agent
swarm agent replay <agent-id> --event-id <id> is the same primitive scoped to one agent.
swarm agent replay-backlog <agent-id> redelivers an agent’s pending backlog — useful after a
restart, if swarm run status shows an agent still blocked on undelivered events.
Crash recovery
On restart, the runtime replays any event that was persisted but not yet marked delivered. Each event carries an idempotency key — a fingerprint the runtime checks before acting — so replaying an already-processed event is a no-op rather than a second execution, and recovery does not double-process work.Run-level fork
Fork reconstructs the full system state at a moment in time and resumes from there. Forks currently run against the same contract bundle as the source run; forking onto a new bundle is planned but not yet enabled.swarm run trace <new-run-id> -f.
Flags:
--at-event <event-id>— fork at this source event; the new run starts as if the source run had just processed this event.--bundle-hash <hash>— pin the fork to a specific bundle. Same-bundle forks only; cross-bundle forks are gated separately and currently rejected.--idempotency-key <key>— retry-safe fork creation; identical key returns the same fork rather than creating a duplicate.