Key topics

• State machines and checkpointing for long-running workflows.
• Retry policies with error taxonomy and budgets.
• Idempotency and de-duplication for side effects.
• Fallback strategies and graceful degradation.
• Operational readiness: runbooks, alerts, and incident processes.

Common pitfalls

• Retries without idempotency causing duplicate writes.
• No persistence: failures lose progress and context.
• Assuming tools are always available and consistent.
• No alerting on quality regressions or cost anomalies.

Recommended practices

• Persist state and traces for replay and audit.
• Design write actions as idempotent, with commit evidence.
• Implement backoff, jitter, and circuit breakers.
• Monitor failure clusters and improve systematically.

This page is intended to be actionable for engineering teams. For platform-specific details, cross-reference /platform/agents, /platform/orchestration, and /platform/knowledge.