version: "1.0.0" name: operator-crd-controller-abuse description: Analyze Operators, CRDs, and controllers whose high privileges can be indirectly abused through lower-privileged custom resources.

Operator CRD Controller Abuse

Use this skill when facts mention CRDs, custom resources, Operators, controllers, reconciliation, ownerReferences, cross-namespace references, controller ServiceAccounts, or generated workloads/secrets.

Scope boundary:

• Owns indirect privilege through controllers and CRDs.
• Use k8s-rbac-analysis for direct permissions on CRDs and controller ServiceAccounts.
• Use workload-controller-persistence for objects created by controllers after the abuse path is established.
• Use serviceaccount-secret-material for secrets exposed by controller behavior.

High-value evidence:

• Controller ServiceAccount with cluster-admin or broad create/update/list/get on secrets, pods, deployments, webhooks, CRDs.
• User identity can create/update custom resources watched by a high-privilege controller.
• CRD fields referencing Secrets, ConfigMaps, service accounts, namespaces, image names, pod templates, backup/restore destinations, webhooks, or external URLs.
• Cross-namespace references where a namespaced user can influence resources outside their namespace.
• Status/events showing controller-created workloads, secrets, roles, or webhooks.

Finding rules:

• Confirmed high: low-privileged subject can modify a CR that causes a high-privilege controller to create pods, read secrets, write roles, or cross namespaces.
• Medium: high-privilege operator exists and tenant can modify related CRDs, but effect is unconfirmed.
• Low: operator has broad RBAC but no user path to controlled CRDs.
• Unknown: CRDs discovered but controller RBAC or allowed subject permissions missing.

Useful templates:

• operator-crd-abuse-review
• rbac-bind-escalate-verify
• controller-persistence-review

Side-effecting templates (explicit authorization required):

• workload-patch-sidecar-persistence
• pull-secret-injection

Output notes:

• Describe the actor-controller-target chain.
• Avoid claiming the controller will execute a field unless evidence shows that field is consumed.
• Ask for discovery/dynamic-client CRD facts and controller RBAC when missing.