openshift-app — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited openshift-app (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Package, build, secure, and deploy applications on OpenShift Container Platform 4.14-4.22. Covers container images, deployment manifests, CI/CD pipelines, security hardening, operational patterns, and disconnected environments.
| Task | Go to |
|---|---|
| Build a container image for OpenShift | Container Images below; references/container-images.md for Podman, ImageStreams, certification |
| Choose Helm vs Kustomize vs Operator | Packaging Decision Matrix below; references/packaging-formats.md for OLM v1 RBAC and certified charts |
| Fix SCC / permission errors | references/security.md (Restricted-v2 section) |
| Set up CI/CD pipeline | references/cicd-gitops.md |
| Harden supply chain (sign, attest, scan) | references/security.md (Supply Chain section) |
| Configure Routes, probes, scaling | references/operations.md |
| Deploy in air-gapped / disconnected env | references/disconnected.md |
| Migrate from DeploymentConfig | references/gotchas.md (DeploymentConfig section) |
| Understand OCP version breaking changes | references/gotchas.md (Version Timeline section) |
OpenShift assigns a random UID from a namespace-specific range but always sets GID 0 (root group). Hardcoded USER 1000 in Dockerfiles will fail under restricted-v2 SCC.
# OpenShift-compatible Dockerfile pattern
FROM registry.access.redhat.com/ubi9/ubi-minimal:latest
COPY --chown=1001:0 app /app
RUN chmod -R g=u /app && \
chgrp -R 0 /app
# Use 1001 as conventional non-root UID
# OpenShift ignores this and assigns its own UID, but vanilla K8s respects it
USER 1001
EXPOSE 8080
ENTRYPOINT ["/app/server"]Key rules:
chgrp -R 0 && chmod -R g=u (mirror owner perms to root group)runAsUser empty in pod spec["binary"] (not shell form) for signal propagationreadOnlyRootFilesystem: trueAll authenticated users get restricted-v2. It is stricter than vanilla K8s PSS restricted:
| Field | restricted-v2 | K8s PSS restricted |
|---|---|---|
| Capabilities | Drop ALL | Drop some |
| allowPrivilegeEscalation | false (enforced) | false |
| seccompProfile | RuntimeDefault required | RuntimeDefault required |
| runAsUser | MustRunAsRange (namespace range) | MustRunAsNonRoot |
| Volume types | configMap, downwardAPI, emptyDir, PVC, projected, secret | Same + ephemeral |
Minimum compliant pod securityContext:
securityContext:
runAsNonRoot: true
# Do NOT set runAsUser -- let OpenShift assign from namespace range
seccompProfile:
type: RuntimeDefault
allowPrivilegeEscalation: false
capabilities:
drop: ["ALL"]Confirm it took effect on a running pod — which SCC admitted it, and the UID actually assigned from the namespace range:
oc get pod NAME -o jsonpath='{.metadata.annotations.openshift\.io/scc}{"\n"}'
oc get pod NAME -o jsonpath='{.spec.containers[0].securityContext}{"\n"}'
oc get pod NAME -o jsonpath='{.spec.securityContext.runAsUser}{"\n"}'An empty runAsUser with a populated SCC annotation is the expected result: the range assignment lands on the pod spec at admission, not in the manifest.
PSA runs in parallel with SCCs. A pod must pass both. OpenShift auto-labels namespaces with PSA levels matching the most privileged SCC available.
Argo CD now bundles Helm 4. PR #28076 ("feat: Migrate from Helm 3 to Helm 4") merged 2026-06-09 and first shipped in v3.5.0 (2026-08-04). The pin, read from hack/tool-versions.sh at each tag:
| Argo CD line | Helm pinned | Note |
|---|---|---|
| 3.5.x (v3.5.3, 2026-09-14 — current latest) | helm4_version=4.2.1 | Helm 4 |
| 3.4.x (v3.4.9, 2026-09-14 — parallel maintenance) | helm3_version=3.19.4 | still Helm 3 |
So "Argo CD only supports Helm 3" is no longer true, and the old "through v3.3" framing understated it: 3.4.x stayed on Helm 3 too, and only 3.5.x jumped.
Upgrading 3.4 → 3.5 can change your rendered manifests with no chart or values change. Helm 4 altered null/nil value coalescing, so explicit null fields can start appearing. Tracked in argoproj/argo-cd#29068, still OPEN — diff a render before and after rather than assuming parity.
for the 4.22 line (Web Terminal Operator 1.17, 2026-06-24, which also bumped oc to v4.22.1); 4.19-4.21 bundled v3.17.1. Sourced from the redhat-developer/web-terminal-operator CHANGELOG, because the Red Hat docs page is a client-rendered shell with no version text in its HTML for any OCP version.
release-policy pages state only that the most recent minor gets fixes, with no Helm 3 sunset date, and Helm 3 is still shipping patches (v3.21.4 on 2026-08-14) alongside the 4.x line. Treat "Helm 3 is dead" as unsupported.
3.5.x is the Helm 4 migration — do it deliberately, with a render diff, not as a side effect of taking the latest tag.
Use Deployment for all new work. For ImageStream triggers on Deployments:
metadata:
annotations:
image.openshift.io/triggers: >-
[{"from":{"kind":"ImageStreamTag","name":"myapp:latest"},
"fieldPath":"spec.template.spec.containers[?(@.name==\"myapp\")].image"}]Also set lookupPolicy.local: true on the ImageStream.
Must migrate to OVN-Kubernetes before upgrading. Key impacts:
100.64.0.0/16 and 100.88.0.0/16 (check for conflicts)All nodes must run cgroup v2 before upgrading. cgroup v2 was the default for new installs since 4.14, deprecated in 4.16.
Elasticsearch, Fluentd, and Kibana are gone. Replaced by LokiStack + Vector + console UI plugin. Migration is NOT in-place -- deploy Loki/Vector in parallel, run both stacks during retention window, then retire Elasticsearch.
| Variant | Size (~compressed) | Package Manager | Use Case |
|---|---|---|---|
ubi9/ubi | ~80 MB | dnf/yum | Builder stages, development |
ubi9/ubi-minimal | ~36 MB | microdnf | Light runtime, need to install packages |
ubi9/ubi-micro | ~12 MB | None | Production runtime (multi-stage required) |
ubi9/ubi-init | ~80 MB | dnf/yum | systemd services (StopSignal: SIGRTMIN+3) |
Recommendation: UBI Micro for production runtime via multi-stage build. UBI Minimal as builder stage. UBI Micro is preferred over scratch because compliance scanners classify scratch images as unrecognizable.
UBI is freely redistributable without a Red Hat subscription.
# Stage 1: Build
FROM registry.access.redhat.com/ubi9/ubi-minimal:latest AS builder
RUN microdnf install -y --setopt=tsflags=nodocs --setopt=install_weak_deps=0 \
golang && microdnf clean all
COPY . /src
WORKDIR /src
RUN CGO_ENABLED=0 go build -o /app/server ./cmd/server
# Stage 2: Runtime
FROM registry.access.redhat.com/ubi9/ubi-micro:latest
COPY --from=builder --chown=1001:0 /app/server /app/server
RUN chmod g=u /app/server
USER 1001
EXPOSE 8080
ENTRYPOINT ["/app/server"]If certifying for the Red Hat Ecosystem Catalog:
name, vendor, version, release, summary, description/licenses with software termsdnf update-minimal --security --sec-severity=Important --sec-severity=Critical)See references/container-images.md for full details.
| Format | When to Use | Limitations |
|---|---|---|
| Helm | Distribute to other teams/customers; values-driven config; OperatorHub Helm operators | Helm 3 only on OCP today; chart-verifier for certification |
| Kustomize | Same-team env overlays (dev/staging/prod); GitOps prerequisite | No templating logic; oc apply -k does NOT support --enable-helm |
| Helm + Kustomize | Dominant hybrid: Helm for packaging, Kustomize for env patches | Requires --enable-helm flag (only works in kustomize build or ArgoCD) |
| Operator (Go) | Stateful apps needing Day-2 ops (backup/restore/scaling); L3-L5 maturity | Complex to develop; Operator SDK CLI deprecation announced at OCP 4.16, 4.18 was the last OpenShift to ship it — on 4.19+ install it yourself from upstream, which continues |
| Operator (Helm) | Simple operators for OperatorHub distribution | Limited to L1-L2 capability maturity |
| OLM v1 ClusterExtension | Install operators on OCP 4.18+ | Requires user-provided ServiceAccount + RBAC; AllNamespaces only |
| OpenShift Templates | Legacy only (NOT recommended for new work) | Not portable to vanilla K8s; Template Service Broker removed in 4.4 |
Detect OpenShift at template time:
{{- define "mychart.isOpenshift" -}}
{{- if .Capabilities.APIVersions.Has "security.openshift.io/v1" -}}true{{- end -}}
{{- end -}}Conditional Route vs Ingress:
{{- if .Capabilities.APIVersions.Has "route.openshift.io/v1" }}
apiVersion: route.openshift.io/v1
kind: Route
{{- else }}
apiVersion: networking.k8s.io/v1
kind: Ingress
{{- end }}SCC-compatible values (let OpenShift assign UIDs):
securityContext:
runAsUser: null # Do NOT hardcode
fsGroup: null # Do NOT hardcode
runAsNonRoot: trueSee references/packaging-formats.md for OLM v1, Kustomize patterns, and certified chart requirements.
| Reference | Contents |
|---|---|
references/container-images.md | UBI variants, multi-stage builds, arbitrary UID, certification, Podman, ImageStreams |
references/packaging-formats.md | Helm on OCP, OLM v1 RBAC, Kustomize overlays, certified operators/charts |
references/security.md | SCC/PSA, supply chain (Sigstore/RHTAS/Conforma), secrets (ESO/Vault), FIPS, compliance, NetworkPolicy |
references/cicd-gitops.md | Tekton Pipelines, Chains, Pipelines-as-Code, ArgoCD Agent, Shipwright, image promotion |
references/operations.md | Routes/TLS, probes, HPA/KEDA/VPA, monitoring, logging, storage, sidecars, serverless, multi-arch |
references/gotchas.md | Version timeline (4.14-4.22), DeploymentConfig migration, SDN removal, networking changes |
references/disconnected.md | oc-mirror v2, registry mirroring, OCI GitOps, Tekton bundles, OSUS upgrades, air-gap patterns |
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.