When OpenClaw engages in multi-step automation (scheduled, event-driven, cross-machine), reliability depends on clear dependency graphs and observability rather than single-point scripting tricks.
1. Task dependency and idempotence
Define input/output contracts for each step; check whether the disk has been partially written before retrying on failure to avoid repeated side effects. Use idempotent keys or deduplication tables for external APIs.
2. Retry and backoff
Use exponential backoff and set an upper limit; immediately fuse and alert for unrecoverable errors (authentication failure, quota exhaustion) instead of blindly retrying to fill the queue.
3. Align with lease term
On MacCloud daily/weekly instances, orchestration tasks should be time-aware for expiration; reserve buffers before long tasks start or run critical paths on subscription instances.
4. Observability baseline
Unified structured log fields (run_idγstepγlatency_ms) to facilitate quick context alignment between column-related articles and work orders.
