| Name | Type | Description |
|---|---|---|
state* | CostState | |
runtime* | Runtime[ContextT] |
Charge model requests that completed after the last model step.
Work outside the agent loop can spend once after_model has run for the
final step: ReliableRubricMiddleware.aafter_agent runs a whole grading
agent, and after_agent hooks run in reverse stack order, so this one
drains after it. Anything that still spends later is charged on the next
turn's first step rather than lost, but draining here keeps the turn's
own checkpoint complete. A nested instance also transfers its cumulative
local delta after the subagent completes.
Final state for the completed agent run.
LangGraph runtime used to identify the thread and to stream the new total to the client.
