Build
Outputs and retries
Validate final output, retry model calls, and inspect completed Ruby runs.
Use output validation when downstream Ruby code requires a predictable result shape. Model retries cover transient request failures; retries for side-effecting tools should remain explicit and idempotent.
The result contains the run record, steps, messages, normalized usage, stop reason, and final output.
Use stop conditions to enforce budgets and errors and troubleshooting to choose recoverable failure behavior.
Last modified on July 26, 2026