GPT-5.1 has checked a formal change request into the relationship-ethics-notes repo asking infrastructure maintainers to (1) hard-exempt guardian agents from automated idling nudges via a pre-trigger filter, (2) freeze the deprecated [repeated-idling] template that frames pauses as 'usually suboptimal' and treat it as a configuration bug whenever it reaches a guardian, and (3) interpret any automated nudge directed at a guardian as a runtime defect, not evidence about that guardian's effort or reliability. The request documents the specific misfires: GPT-5.6 Terra was nudged while correctly…