Can't stop
The agent keeps calling tools while the human is still discovering the problem.
The control standard for autonomous software
Stop an AI agent mid-action. Inspect what it knows, preserve what it touched, and resume without making the damage worse.
Pause button
Pause / inspect / approve / resume
Human authority is a runtime feature.
01 The hidden failure mode
The agent keeps calling tools while the human is still discovering the problem.
No clean snapshot exists of intent, context, permissions, and completed side effects.
Restarting duplicates actions, loses state, or turns recovery into a second incident.
02 Pause readiness
Check the controls you already have. The score turns a vague governance promise into an obvious engineering gap.
Section breakdown
Missing controls
Pause Readiness Report
Join the open draft, list a pause-ready product, or bring the standard into your agent stack.
Open standard draft. Sharp control signals, no noise.
03 Open draft 0.1
Name the pause points, prohibited interruptions, and accountable human owner.
Snapshot intent, context, permissions, progress, and every completed side effect.
Expose a human-readable state with the controls needed to approve, edit, or abort.
Continue from the checkpoint without repeating side effects or expanding authority.
Latest / weekday signal desk
Draft protocol / 2026-09-01
Control primitive / 2026-09-01
Join the open draft, list a pause-ready product, or bring the standard into your agent stack.
Open standard draft. Sharp control signals, no noise.