pi-workflows

You orchestrate multi-step agent workflows by defining them in .workflows/*.workflow.yaml files. The system validates your workflow specs and executes them using DAG-based orchestration before running the pipeline.

DAG-based execution structures agent pipelines as a directed acyclic graph, ensuring typed data flows strictly between steps. This mechanism enforces strict output schemas to maintain data integrity throughout the workflow.

Yes, you can resume a failed agent workflow from a specific checkpoint. The checkpoint and resume mechanism captures execution state, allowing you to debug step failures and continue the pipeline without restarting.

You validate workflow specs and enforce output schemas before execution by applying the built-in validation mechanism. This checks your YAML configurations and ensures typed data flow compliance prior to starting the pipeline.

The best way to debug step failures in complex agent pipelines is by inspecting agent configurations and utilizing checkpoint and resume functionality. This isolates failing steps and validates data flow without rerunning the entire workflow.