An internal gate an agent runs before taking an unprompted or proactive action — e.g., interjecting in a Slack thread — asking itself whether it should act now rather than acting by default whenever it technically could. Linear's agent uses this to avoid over-eager, unwanted interruptions; one illustrative failure mode surfaced in testing was a user calling the agent 'dude' and the agent refusing to respond because the tone read as insufficiently formal — a sign the check itself needs tuning, not that the pattern is wrong. Related to human-attribution-accountability-pattern: both concern when and how the agent should insert itself into a human's workflow.