OpAgent Runtime is the execution and permission authority for the active local or remote workspace. The portable enforcement design lives in the OpAgent sandbox documentation. This page records OpenBrain's product-facing policy.
OpenBrain presents Ask for approval, Approve for me, Full access, and Custom profiles. The Runtime stores an immutable permission snapshot on each thread; subagents inherit it as a ceiling. A model, repository, Skill, remote Tool, or scheduled task cannot select or broaden its own profile.
Full access requires explicit user confirmation and trusted workspace state. Safe profiles fail closed when the platform sandbox is unavailable. Approval is pre-action: a later diff or warning cannot undo an external side effect.
“Approve for me” uses the product's trusted ApprovalReviewer hook. If review
is unavailable, times out, or fails validation, the request returns to user
review or is denied; it is never silently approved. The public Runtime does not
contain a product reviewer or product account implementation.
OpenBrain session resolution and managed-service authorization are compiled in
integrations/opagent. Unknown system services and
remote nodes fail closed. User configuration cannot grant private
openbrain-server trust or turn an arbitrary remote manifest into the managed
GBrain endpoint.
Local and remote workspaces follow the same rule: enforcement happens on the machine that owns and executes the workspace Runtime.