authority-signoff-call

Authority Sign-Off Call

Autonomous agents that simulate or automate incident-response and government-operations decisions — dispatch, budget release, disaster-management sanctions — are often designed not to wait for a human before acting, since the whole point is that the workflow doesn't stall on a dashboard nobody's watching. But someone real is still accountable for what got authorized in their name, and a push notification they might never open doesn't close that loop. A phone call does.

This skill is exactly that call: it reaches the specific named official a decision was auto-authorized under — on their own pre-registered line, wherever they actually are — reads the decision out loud, and returns a real, attributable confirm or override. It is one accountability primitive, not a certified safety mechanism — see "Rules you must follow" below for the hard limit on what the returned decision may be used for.

It does not invent a decision-application mechanism. It feeds the structured outcome into whatever function your system already uses to apply a human decision to that record. Two integration surfaces are documented below — use whichever matches where the decision to place the call is actually happening.

When to use

• Your system auto-authorizes actions under a real authority model (a policy tier, a delegated-signatory rule, an on-call-independent role like "budget owner" or "duty magistrate") and the record of who it was authorized as names a specific real person or role.
• That person is not at a keyboard when the decision gets made — the whole point of autonomous authorization is that it doesn't wait for them — but a phone call can still reach them for the accountability step afterward.
• You want the confirm/override to be a real spoken exchange with the named authority, not a push notification that "was sent."

When not to use

• The action hasn't happened yet and you need to block it until a person says yes first. Use deployment-approval-call instead — that skill is a pre-action gate with code-verified approval.
• You need an on-call ladder — dialing through a list of people until one of them commits to owning an incident. Use incident-escalation-call instead.
• You don't have a specific, pre-registered phone number that the accountable person themselves configured. Never guess a number, resolve one from a directory, or call someone else "on their behalf."
• The decision is medical, legal, or a real emergency dispatch. This skill calls a named accountable person about a decision their own system already made — it is not a dispatch, notification-of-record, or emergency-services mechanism.
• You want a second attempt at a different answer after one call already resolved. One call, one attributable answer.