cloudflare-workflows

Long-running workflows in Cloudflare Workers use durable execution to persist state between steps, automatically retry failed operations with exponential backoff, and continue processing after interruptions. This Skill provides step.do, step.sleep, and event-driven waits to orchestrate multi-step tasks lasting hours to days without losing progress.

Yes. Event-driven workflows let you pause execution and wait for external events—webhooks, approvals, user actions—using step.waitForEvent. The workflow persists its state until the event arrives, then resumes automatically, eliminating polling loops and timeout complexity.

Durable workflows apply automatic retries with configurable exponential backoff to failed steps while preserving state between attempts. You can schedule delays with step.sleep and step.sleepUntil, set timeouts, and raise NonRetryableError to stop retrying—all without manual retry logic in your code.

Use wrangler to deploy workflows written in TypeScript with @cloudflare/workers-types and WorkflowEntrypoint integration. Wrangler handles local testing and deployment patterns; ensure all step returns are JSON-serializable to satisfy durable execution constraints.

Yes. This Skill handles cross-API coordination by orchestrating multiple dependent API calls with retries and state persistence. Common use cases include payment processing with automatic retry logic, waiting for confirmation webhooks, and gracefully handling timeouts across distributed systems.

Step return values must be JSON-serializable; complex objects require custom serialization. Workflows run within Cloudflare Workers' resource constraints and timeout windows. NonRetryableError prevents retries for specific failures, but you must explicitly handle edge cases in business logic.