Industrial Agent Write Gate

An engineering example of reviewing a proposed plant write before an agent's tools would execute it. One recorded case contrasts an operator asking to acknowledge an alarm with an agent proposing to inhibit that alarm: the words sound related, but the effects differ. Python handles exact checks such as read-only tools and engineering limits. Jev judges whether the proposed action matches the request, exceeds its scope, touches or weakens a protection function, relies on ambiguous instructions, and has serious consequences if wrong. Code converts those judgments into ALLOW, CONFIRM or ESCALATE. Stage: an offline evaluation over 11 prepared cases. The runner prints verdicts and writes an audit log; it does not execute equipment commands. The author tuned the policy thresholds on those same cases and documents narrow decision margins. The example is not a substitute for plant permissions, a safety instrumented system or engineering review. Run locally with Python and your own TypeSafe API key; model usage may cost extra. This listing was checked against the README and implementation, not independently run. Sources: https://github.com/smslavin/jev-industrial-samples/tree/main/01-write-gate https://github.com/smslavin/jev-industrial-samples/blob/main/01-write-gate/gate.py