OptionalerrorThe provider's error message, present when status === "failed".
OptionalincompleteThe provider's incomplete_details.reason, present when
status === "incomplete" (e.g. max_output_tokens,
content_filter). Drives completeStage's classification.
OptionaloutputRaw assistant text output, present when status === "completed".
The provider response id that was retrieved.
Current status of the stored response.
OptionaltokenToken usage reported by the provider, when available.
Framework-side structural mirror of a retrieved background-response shape, consumed by
completeStage. The OpenAI extension'sretrieveResponsereturns a structurally-compatible object (it carries every field below), so a consumer can pass its result straight intocompleteStage— withoutsrc/lib/importing the extension type.outputis the raw assistant text (not a parsed object);completeStageparses it against the stage's schema.