Observable Workflow Design

Use this skill for workflows where "the request succeeded" is not the same as "the work finished." Typical examples: account provisioning, runtime setup, data import, repo indexing, payment-to-entitlement activation, migration jobs, deployment pipelines, webhook handling, and cross-service sync.

This skill complements production-observability. That broader skill covers the general telemetry stack; this one turns one workflow into a debuggable state machine.

Iron Rule

Do not patch a failing multi-step workflow until you can name the failed step and show the evidence. If two patch-test cycles fail, stop and add/read workflow diagnostics before changing behavior again.

Workflow Contract