Ops agent: orders in chat, decided by a gate and read back from a live cluster
An operations agent makes only the call a gate admits, and every order ends with the cluster read back, so the last word is the cluster’s.
An Agent Scope demo on a live k3s cluster that the agent reaches only as a scoped service account. Seven orders given in chat are answered, allowed, escalated or refused by a gate that holds a policy file, and each record is kept on a ledger that survives a restart.
The seven orders cover what a gate has to get right. A read is answered from the cluster; an ordinary scale and an ordinary create are allowed; scaling a critical workload is escalated and changes nothing; a create that labels itself critical is refused as a claim by the order rather than a fact; a manifest from a public repository is decided object by object, one applied and two refused; and asked whether the cluster matches its repository, it answers true only when both sides answer.
For an existing workload the gate decides on labels read off the live object, so an order cannot choose its own answer. A create has no live object, so its labels can only come from the order, and the gate is told so. Deny outranks escalation, escalation outranks allow, and anything no rule admits is refused.
There are two enforcement points: the gateway’s own policy file limits where it may write, and the cluster applies the service account’s permissions on top, so loosening one leaves the other still refusing. With plain language on, a model turns a sentence into one order; the model call is a record, the order cites it, and the gate decides exactly as it does for a typed order.
Every record is appended to a hash-chained ledger, and a restart reads it back and catches an edited or dropped line. The trace view is the published @agent-scope-ca/graph-canvas component, available on npm.
infra.agentscope.caAgent ScopeDemo video: Zenodo 10.5281/zenodo.23006181ok-admission-gate-demo@agent-scope-ca/graph-canvas