Writing Agent Relay Workflows

⚠️ DEPRECATED — this is the v1 engine

This skill documents Relayflows v1: @relayflows/core's WorkflowBuilder, the chained builder (workflow('name').pattern('dag').agent(...).step(...).run()) with failOnError, captureOutput, {{steps.X.output}} templating and .onError(). v1 is superseded.

New work goes to the v2 engine — @relayflows/surface's flow() and the YAML/JSON dialect compiled by @relayflows/sdk, CLI binary flows, packages 2.0.x. Read writing-relayflows for the authoring surface and relay-80-100-workflow for the v2 validation shape. None of failOnError, captureOutput, {{steps.X.output}}, .onError() or idleThresholdSecs exists in v2, so following this skill for a v2 flow produces something that will not compile.

Everything below is kept intact for people reading or maintaining an existing v1 workflow.