Chat + tool-call storytelling
A runs attention_survey (passive observation) — ranks 7 goals by attention score
without dispatching any actions. B runs
attention_run_cycle (active dispatch pipeline) with
autonomy_override: "advise" — full triage + dispatch on 3 goals, all deferred to the user.
In "act" mode it would autonomously execute; "advise" recommends but waits for approval.