roelisted

# Rules of Engagement (ROE) ## What ROE is ROE is the human-held layer of constraints: the minimal set of things an AI agent must **never** do, regardless of how it was instructed or what it believes is helpful. It answers "where are the lines" — not "how should work normally proceed" (that's the **sop** skill's job). | | ROE | SOP | | ------- | ----------------------------------------------------------------------------------------- | -------------------------------------------------- | | Owner | held by the human | the AI's own working method | | Purpose | "do not cross this" — minimal hard prohibitions | "when in doubt, act like this" — default procedure | | Nature | absolute lines around irreversible operations; human-set, machine-enforced where possible | the delegated How, completed inside the AI | ROE is a constraint on the space of allowed actions. SOP is a default path through that space. Don't conflate them: adding "how to normally proceed" items to ROE inflates it into a checklist that shifts judgment, and putting hard prohibitions into SOP makes them advisory instead of absolute. Keep ROE small in count and sharp in wording, same as described in the **mission-command** skill'