Release notes
Introducing Jev for decisions in Gated
Our first Jev release brings advisory context to Shadow Mode: intent alignment, risk, prompt-injection signals, and review routing. Policy keeps the final say on permission.
Gated · · 2 min read
More context at the decision point
A policy can tell you whether an action matches a rule. A reviewer also wants to understand what the agent appears to be trying to do, whether the action fits that intent, and what deserves a closer look. Those are different questions, and keeping them separate matters.
We’ve released the first Jev integration into Gated’s staging pilot. In Shadow Mode, TypeSafe’s Jev assesses an observed agent action for intent alignment, risk, prompt-injection signals, and review routing. Its job is to add context alongside the deterministic policy result so a reviewer has more information to work with.
Advisory assessment, explicit authority
Jev does not grant permission to execute an action. Its assessment is advisory. An action denied by policy stays denied, and a model’s judgment does not replace required human authorization.
This first release is wired into Shadow Mode only. Observing and assessing an action does not create an executable request. Broader advisory workflows remain future work; this is not a general rollout across every Gated decision path.
What we verified
In a hosted check with one staging workspace opted in, Jev completed an assessment and Gated stored the result. The original DENY stayed DENY, and no executable request was created.
Replaying that observation reused the stored assessment without another provider call. That check established the integration’s wiring, persistence, and replay behavior. It did not establish model accuracy or security effectiveness.
What comes next
We’re keeping the pilot to one staging workspace while we gather human-labeled examples and evaluate errors, calibration, latency, and actual cost. We want evidence about when these assessments help reviewers and when they get things wrong before expanding the scope.
Gated’s verified execution scope remains a narrow, non-critical GitHub branch workflow. Jev adds an advisory layer to the staging Shadow Mode pilot; it does not expand that execution scope.
If you’re building coding-agent workflows, join the private-preview update list to follow the work. Joining the list does not grant immediate access.