@x12i/xentra-approvals

Human-in-the-loop approval policies and requests for sensitive agent or user actions.

Install

npm install @x12i/xentra-approvals

Requires @x12i/xentra-store.

Usage

import { createMemoryStore } from "@x12i/xentra-store";
import { createApprovalsService } from "@x12i/xentra-approvals";

const store = createMemoryStore();
const approvals = createApprovalsService(store);

const policy = await approvals.createPolicy({
  accountId: "acc_1",
  name: "High-risk tool use",
  actorTypes: ["agent"],
  actions: ["tools.invoke"],
  resourceTypes: ["tool"],
  requiredApproverRoleIds: ["role_acc_1_security-admin"],
});

const request = await approvals.createRequest({
  accountId: "acc_1",
  actor: { type: "agent", id: "agent_1" },
  action: "tools.invoke",
  resource: { resourceType: "tool", resourceId: "tool_1" },
  requestedByActor: { type: "agent", id: "agent_1" },
  reason: "Sending customer email",
});

await approvals.approve("acc_1", request.id, "user_approver");
// or approvals.reject("acc_1", request.id, "user_approver", "Not justified");

Policy matching

matchesPolicy(request, policies) returns whether any active policy applies to a proposed action (by actor type, action, and resource type).

Related packages

• @x12i/xentra-permissions — can return requires_human_approval when wired
• @x12i/xentra-audit — approval events are written through the store
• @x12i/xentra-react — ApprovalInbox component scaffold

Development

npm run build -w @x12i/xentra-approvals
npm test -w @x12i/xentra-approvals

License

MIT