Skip to main content
No. The registry stores, indexes, and serves packages; the guardrails service statically scans files and never executes anything either. Running a skill is the caller’s own runtime’s responsibility.
No — Level 1 is mandatory and enforced by the guardrails service itself, not by caller configuration. Only Level 2–4 checks are configurable, and only Level 2 is on by default.
They’re recorded as guardrail_warning_ids on the immutable publish audit event, not in a separate scan-result store — a tenant audits warning trends from the audit log itself.
So the API never reveals that a private skill exists to someone who can’t see it. A 403 would leak existence; a 404 looks identical to “there’s nothing here.”
No — every rule, catalog or custom, must be evaluable fully offline from the package’s own files. This is a hard design constraint of the guardrails service, not a current limitation.