Agno now saves failed workflow runs before the stream ends

A failing Agno workflow run now writes ERROR to the database first, then sends StepError and WorkflowError. Before, a streaming run that failed halfway emitted its WorkflowError first and saved the failure after it, so a client that stopped reading at that event could lose the run entirely.

A workflow that resumed after a pause could also fail, save ERROR to the database, and still leave its tracked stream stuck at PAUSED.

The outputs of steps that completed before the error stay on the run.

run = workflow.get_last_run_output()
run.status                                         # RunStatus.error
[(s.step_name, s.success) for s in run.step_results]
# [("fetch_orders", True), ("charge_cards", False)]

Agno also stopped retrying guardrail rejections in nested workflows. Those retries had replayed the side effects of earlier child steps.

Learn more about workflow sessions, and the run records they store, in the documentation.