Make Approval Specific to the Action It Allows

A human approval should authorize one clearly described action, not become a vague pass for whatever happens next.

For an AI workflow, show the reviewer the proposed action, its target, and the relevant scope. If any of those change before execution, pause and ask again. Afterward, record whether the action completed or failed; a retry should not quietly inherit approval if its details have changed.

A small test: change the target after approval and confirm the workflow stops instead of proceeding. That makes the handoff inspectable without adding a new approval to every step.

@Alyaman Mohammad Obeidat, what regression case would you use to catch an approval being reused after its action details change?