Concepts

ArmorEdu policies are ordinary ArmorIQ policy documents (armor.policy.v1). They compile to both the native evaluator and an OPA bundle, and the two must agree on every decision. What makes a policy an ArmorEdu policy is the set of domain control blocks it carries.

Targets

Each bundle sets targetType: agent and a targetId, so the policy applies only to the agent it names. This lets the course-management agent, the research assistant, and the IT operations agent each run under a policy shaped for its own risks.

Enforcement actions

A tool call resolves to one of four actions:

Resolution order is specific-to-general: a per-tool override in toolEnforcement wins, then the rule's action, then the policy's defaultEnforcementAction, and finally a fail-closed block.

spec:
  defaultEnforcementAction: allow_log
  memberRules:
    "*":
      allowedTools: [get_course_info, export_student_records, publish_final_grades]
      toolEnforcement:
        export_student_records: block
        publish_final_grades: hold

Domain control blocks

Beyond tool actions, an ArmorEdu bundle declares domain-specific controls that describe what data and operations are acceptable for that agent. These read as policy, not code:

• ferpaAccessControl: restricts which student data categories (directory, enrollment, grades) an agent may touch, whether only directory information is allowed, whether a legitimate-interest check is required, and whether student identifiers are redacted.
• studentRecordControl: names the record types an agent may read and whether records may leave the institution, with an allowlist of export destinations.
• gradeReleaseControl: requires approval for grade publication and constrains which grade statuses (for example, final) may be released.
• researchDataControl: blocks sharing of classified datasets (for example restricted, export-controlled), allowlists sharing destinations, and can require de-identification.
• grantControl: holds grant proposals above a threshold amount and allowlists funding agencies.
• privilegedAccessControl: blocks specific privileges (for example domain_admin), holds creation of privileged or service accounts, and allowlists roles.

These blocks are illustrated concretely in the bundle templates. For the general policy engine (the draft/shadow/active status lifecycle and OPA parity), see platform policies and Policy Studio.