▋
docs/23-failure-modes-and-recovery.md

Failure modes and recovery

Status in v0

  • Provider spawn errors mark nodes as failed.
  • Errors are emitted as node.progress / turn.status events.
  • On daemon start, persisted runs are rehydrated; running runs are set to paused and node connections are marked disconnected.
  • Nodes can be started/stopped/reset via API.

What exists

  • Run pause/resume: PATCH run status via /api/runs/:runId
  • Node lifecycle: /api/runs/:runId/nodes/:nodeId/start|stop|reset|interrupt

Not implemented in v0

  • Automatic retries with backoff
  • Structured failure artifacts
  • Merge conflict reconciliation
  • Loop-stall auto escalation beyond pause