No model-generated shell
Validator commands come from trusted adapters. Model text is never treated as an executable command.
SECURITY BY ARCHITECTURE
Code stays under explicit authority.
Verify is designed around J OS' capability model: intelligence can inspect broadly only where permission exists, while execution authority remains narrow, explicit and auditable.
Untrusted code is sandboxed
Untrusted projects require isolated execution. No silent fallback to host execution.
Read and execute are separate
Project reads and validator execution require distinct scoped capabilities.
Scrubbed environments
Validation processes do not inherit J's full secret-bearing environment.
Private enterprise deployment
JVerify can run inside an institution's own infrastructure so source code and project information do not need to leave its security perimeter.
Auditable decisions
Authorization and validation outcomes can be surfaced through J OS audit and event facilities.