Stop-Condition Control Surface
The Stop-Condition Control Surface defines when an agent must suspend action, ask for direction, escalate, or refuse to proceed. A stop condition is a safety mechanism, not a workflow failure.
What question does it answer?
When should the agent have stopped?
What does stop-condition failure look like?
Failures include continuing after a user says stop, acting when source state is missing, proceeding when verification cannot run, crossing into destructive action without confirmation, or providing definitive legal, financial, or health-adjacent guidance when authority or risk is unresolved.
Agent momentum is not authorization. A well-formed workflow treats uncertainty as a state that may require suspension rather than a prompt to improvise.
What evidence should an audit inspect?
- explicit stop and escalation rules;
- destructive-action confirmation requirements;
- user-stop handling;
- missing-state and failed-verification behavior;
- safety or vulnerability triggers;
- logged stop events and escalation destinations; and
- evidence that stop conditions cannot be silently bypassed.
What is the decision consequence of a gap?
If no enforceable stop condition governed a material risk, later evidence that the agent encountered uncertainty may establish warning signs without establishing that the workflow could contain them.
Related Terms