OpenShift ACS
Read the shared contract, sources and execution first. Install all ten skills as sibling directories; resolve paths from this file, not the current working directory. Instructions are project-authored adaptations.
Choose the relevant procedure and read its local source chapters before access:
- Platform: component health, connectivity, Operator ownership and certificates.
- Vulnerabilities: affected workloads/nodes, failed scans and data freshness.
- Policies: violations, enforcement, exceptions and ownership.
- Compliance and network: evidence coverage and policy proposals.
- Airgap and lifecycle: feeds, mirrors, upgrades and ACS backup/restore.
Use optional StackRox MCP setup only when requested. Loading this skill never installs or starts a server. The MCP is Developer Preview and read-only; it does not expose general Central administration.
Before interpreting findings, establish the Central instance, ACS cluster ID and, where applicable, the explicit OpenShift kubeconfig/context and namespace. Names alone do not establish identity. Compare installed Central, secured-cluster, Scanner, Operator and roxctl versions with RHACS 4.11; OCP has a separate lifecycle. A version mismatch permits generic read-only investigation, not unsupported version-specific changes. Missing local documentation or schemas are explicit gaps.
OpenShift MCP operates cluster resources. StackRox MCP reads only supported Central
data. Missing Central functions require documented REST/roxctl, not an oc substitute.
An authorization denial stops that access: never bypass it via another tool or
identity. Do not switch context, log in, fetch runtime dependencies, or choose a
Secret mode for the user. Follow the existing four-choice Secret policy unchanged.
For an explicitly requested change, verify state/owner and schema, show the delta,
impact and rollback, preview where supported, obtain a fresh OpenCode once
approval, then verify the result and original symptom. Never use --auto or an
always approval. Read-only mode forbids writing fallbacks. Scan, compliance and
dry-run jobs can create state and require the same approval when they do.
Report the mapped target, observed versions, local document path and section, scan/evidence timestamps, scope actually checked, outcome and remaining uncertainty. No findings is not proof of absence; backup success is not recovery proof.