AgentDefinition::$humanReview for a general checkpoint. Approval-gated and manual tools populate $result->run->pause->pendingToolCalls with exact call IDs.
Continue through Agent::continueRun(...) with:
ToolDecisionvalues in approvals or rejectionsToolOutputvalues for manual work- human input for a general review pause
continueRunById(...) with a StateAccessor when the application should reload the run from storage. Decisions are matched by call ID, not tool name.