OptionalAgentOptional parent MJ: AI Agent Runs id for observability. When set (with AgentRunStepID), an
MJ-aware goal engine links the run's prompt runs to this agent run and nests them under the step.
OptionalAgentOptional parent MJ: AI Agent Run Steps id (the goal's step). When set (with AgentRunID), an
MJ-aware goal engine nests a child Prompt step per prompt under it — grouping the goal's many prompt
runs beneath a single step in the realtime agent run.
OptionalContextModel-blind context object. Values are referenced by {{path}} label in the goal/actions and
injected at the action-execution boundary — never seen by any model. (Wired in a later phase; see
plans/realtime/computer-use-remote-browser-blend.md §4.)
OptionalContextThe MJ user the goal run executes as. The CDP layer forwards it to the bound goal engine so an
MJ-aware engine (e.g. MJComputerUseEngine) runs its controller/judge prompts under this user
(prompt-run logging, model access, credential resolution). Transport-neutral here — typed only as
UserInfo, carrying no computer-use SDK types.
OptionalControllerOptional controller (vision/action) model id, when overriding the engine's auto-selection.
OptionalJudgeOptional judge model id, when overriding the engine's auto-selection.
OptionalMaxMaximum perceive-act steps before the loop gives up.
OptionalOnInvoked per step so the caller (e.g. a realtime voice session) can narrate progress.
OptionalSignalAbort signal — when aborted (barge-in), the goal loop stops cooperatively.
OptionalStartOptional URL to navigate to before the goal loop begins.
Options for IRemoteBrowserSession.RunComputerUseGoal. Transport-neutral — concrete model selection (vision/action controller, judge) is resolved by the goal engine the CDP layer binds, so this base type carries no computer-use SDK types.