The problem
Agents can produce valid-looking tool calls that are still unsafe, expensive, or out of scope. A JSON schema validates shape, not intent.
The Jev question
Ask Jev whether the action is allowed, whether it needs confirmation, and how risky it is. Use a noul threshold that matches the cost of a false positive and false negative.
The production loop
Store the Jev decision beside the tool result. Review rejected and approved calls so your policy improves with real workflow outcomes.