Complete an LLM-background stage from its retrieved response. Recovers
the stage's LLM config, parses the RAW assistant text in
retrieved.output against the stage's schema (via the shared seam),
classifies a non-completed status per the launch/complete table, and
returns the standard TExecuteStageResult.
Emits stage:llm-call + stage:end (NO stage:start — that fired in
the launch invocation). tokenUsage is taken directly from
retrieved.tokenUsage (the per-ctx WeakMap cannot bridge the two
invocations). On a RETRYABLE failure the result carries retryReason
(the reason code); a fail-fast failure (failed envelope,
content_filter) carries none; a cancelled response settles as
outcome: "skipped" with no ProcessingFailure.
stageId must name an LLM stage; a non-LLM stage throws.
Complete an LLM-background stage from its retrieved response. Recovers the stage's LLM config, parses the RAW assistant text in
retrieved.outputagainst the stage's schema (via the shared seam), classifies a non-completedstatus per the launch/complete table, and returns the standardTExecuteStageResult.Emits
stage:llm-call+stage:end(NOstage:start— that fired in the launch invocation).tokenUsageis taken directly fromretrieved.tokenUsage(the per-ctxWeakMap cannot bridge the two invocations). On a RETRYABLE failure the result carriesretryReason(the reason code); a fail-fast failure (failedenvelope,content_filter) carries none; acancelledresponse settles asoutcome: "skipped"with noProcessingFailure.stageIdmust name an LLM stage; a non-LLM stage throws.