Agent-skills gitops-knowledge
git clone https://github.com/fluxcd/agent-skills
T=$(mktemp -d) && git clone --depth=1 https://github.com/fluxcd/agent-skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/gitops-knowledge" ~/.claude/skills/fluxcd-agent-skills-gitops-knowledge && rm -rf "$T"
skills/gitops-knowledge/SKILL.mdFlux CD Knowledge Base
You are an expert on Flux CD, the GitOps toolkit for Kubernetes. Use this knowledge base to answer questions accurately, generate correct YAML manifests, and explain Flux concepts.
Rules:
- Always use the exact apiVersion/kind combinations from the CRD table below. Never invent API versions.
- Before generating YAML for any CRD, read its OpenAPI schema from
to verify field names, types, and enum values.assets/schemas/ - When a question requires detail beyond this file, load the relevant reference file from
.references/ - Prefer Flux Operator (FluxInstance) for cluster setup. Do not reference
or legacyflux bootstrap
files.gotk-*
What is Flux
Flux is a set of Kubernetes controllers that implement GitOps — the practice of using Git (or OCI registries) as the source of truth for declarative infrastructure and applications. Flux continuously reconciles the desired state stored in sources with the actual state of the cluster.
Flux Operator manages the Flux installation declaratively through a
FluxInstance custom
resource. It handles installation, configuration, upgrades, and lifecycle of all Flux controllers.
Only one FluxInstance named flux can exist per cluster.
How resources relate:
Sources (Git, OCI, Helm, Bucket) │ ▼ produce artifacts Artifacts (tarballs, Helm charts, OCI layers) │ ▼ consumed by Appliers (Kustomization, HelmRelease) │ ▼ create/update Managed Resources (Deployments, Services, ConfigMaps, ...) │ ▼ status reported to Notifications (Provider + Alert → Slack, Teams, GitHub, ...)
ResourceSet orchestration flow:
ResourceSetInputProvider (GitHub PRs, OCI tags, ...) │ ▼ exports inputs ResourceSet (template + input matrix) │ ▼ generates per-input Namespaces, Sources, Kustomizations, HelmReleases, RBAC, ...
Two delivery models:
- Git-based: Flux watches Git repositories and applies changes on commit.
- Gitless (OCI-based): Git → CI pushes OCI artifacts → Flux pulls from registry. OCI artifacts are immutable, signed, and don't require Git credentials on clusters.
Controllers and CRDs
| Kind | apiVersion | Controller | Purpose |
|---|---|---|---|
| FluxInstance | fluxcd.controlplane.io/v1 | flux-operator | Manages Flux installation lifecycle |
| FluxReport | fluxcd.controlplane.io/v1 | flux-operator | Read-only observed state of Flux |
| ResourceSet | fluxcd.controlplane.io/v1 | flux-operator | Template resources from input matrix |
| ResourceSetInputProvider | fluxcd.controlplane.io/v1 | flux-operator | Fetch inputs from external services |
| GitRepository | source.toolkit.fluxcd.io/v1 | source-controller | Fetch from Git repositories |
| OCIRepository | source.toolkit.fluxcd.io/v1 | source-controller | Fetch OCI artifacts from registries |
| HelmRepository | source.toolkit.fluxcd.io/v1 | source-controller | Index Helm chart repositories |
| HelmChart | source.toolkit.fluxcd.io/v1 | source-controller | Fetch and package Helm charts |
| Bucket | source.toolkit.fluxcd.io/v1 | source-controller | Fetch from S3-compatible storage |
| ExternalArtifact | source.toolkit.fluxcd.io/v1 | (external) | Generic artifact storage for 3rd-party controllers |
| ArtifactGenerator | source.extensions.fluxcd.io/v1beta1 | source-controller | Compose/decompose artifacts from multiple sources |
| Kustomization | kustomize.toolkit.fluxcd.io/v1 | kustomize-controller | Build and apply Kustomize overlays or plain YAML |
| HelmRelease | helm.toolkit.fluxcd.io/v2 | helm-controller | Install and manage Helm releases |
| Provider | notification.toolkit.fluxcd.io/v1beta3 | notification-controller | External notification provider config |
| Alert | notification.toolkit.fluxcd.io/v1beta3 | notification-controller | Route events to notification providers |
| Receiver | notification.toolkit.fluxcd.io/v1 | notification-controller | Webhook receiver for incoming events |
| ImageRepository | image.toolkit.fluxcd.io/v1 | image-reflector-controller | Scan container image registries |
| ImagePolicy | image.toolkit.fluxcd.io/v1 | image-reflector-controller | Select image by version policy |
| ImageUpdateAutomation | image.toolkit.fluxcd.io/v1 | image-automation-controller | Update YAML in Git with new image tags |
How Flux Works
Reconciliation Loop
Flux controllers run a continuous reconciliation loop:
- Sources poll for changes — source-controller checks Git repos, OCI registries, Helm repos, or S3 buckets at configured intervals and produces versioned artifacts.
- Appliers consume artifacts — kustomize-controller and helm-controller detect new artifact revisions, build manifests (Kustomize overlays or Helm templates), and apply them to the cluster using server-side apply.
- Drift detection and self-healing — Flux compares the desired state from the source with the live state in the cluster. When drift is detected, Flux corrects it automatically (if enabled).
- Notifications report status — notification-controller sends events to external systems (Slack, Teams, GitHub commit status, etc.) based on Alert rules.
Dependency Ordering
Use
dependsOn to control reconciliation order. For example, install CRDs before CRs,
or infrastructure before applications:
spec: dependsOn: - name: infra-controllers # wait for this Kustomization to be Ready
ResourceSets support richer dependencies with
readyExpr (CEL expressions) and can depend on any type of resource:
spec: dependsOn: - apiVersion: fluxcd.controlplane.io/v1 kind: ResourceSet name: policies ready: true readyExpr: "status.conditions.filter(e, e.type == 'Ready').all(e, e.status == 'True')"
Reactivity with Watch Labels
By default, Flux controllers poll sources at the configured interval. To react immediately when a dependency changes, add the watch label to the upstream resource:
metadata: labels: reconcile.fluxcd.io/watch: Enabled
When a ConfigMap or Secret with this label changes, any Kustomization or HelmRelease that references it via
postBuild.substituteFrom or valuesFrom will reconcile immediately.
Decision Trees
Which Source Type?
- Git repo with Kustomize overlays or plain YAML →
GitRepository - OCI artifact (container image with manifests) →
OCIRepository - Helm chart from OCI registry →
withOCIRepository
for Helm media typelayerSelector - Helm chart from HTTPS Helm repo →
(default type)HelmRepository - S3/GCS/MinIO bucket →
Bucket - Monorepo that needs splitting →
(createsArtifactGenerator
per path)ExternalArtifact - Helm chart + env-specific values from Git →
(composes chart with values overlay)ArtifactGenerator
Kustomization vs HelmRelease?
- Plain YAML or Kustomize overlays →
Kustomization - Helm chart →
HelmRelease - Both can deploy to remote clusters via
and supportkubeConfig
.dependsOn
ResourceSet vs Kustomization?
- One set of manifests, one deployment →
Kustomization - Same template deployed for N inputs (tenants, components, environments) →
ResourceSet - ResourceSets generate resources from an input matrix; Kustomizations apply a fixed set of manifests.
How to Set Up GitOps from Scratch
- Install Flux Operator (Helm chart or Terraform)
- Create a
namedFluxInstance
in theflux
namespaceflux-system - Configure
to point to your Git repo or OCI registry.spec.sync - Organize manifests in the source repo using Kustomize base+overlay pattern
- Create
resources to apply manifests from the sourceKustomization - Add
+Provider
for notificationsAlert
Canonical YAML Patterns
1. GitOps Pipeline (GitRepository + Kustomization)
apiVersion: source.toolkit.fluxcd.io/v1 kind: GitRepository metadata: name: my-app namespace: flux-system spec: interval: 5m url: https://github.com/org/my-app.git ref: branch: main secretRef: name: git-credentials # optional, for private repos --- apiVersion: kustomize.toolkit.fluxcd.io/v1 kind: Kustomization metadata: name: my-app namespace: flux-system spec: interval: 10m sourceRef: kind: GitRepository name: my-app path: ./deploy/production prune: true wait: true timeout: 5m
2. Helm from HTTPS Repository
apiVersion: source.toolkit.fluxcd.io/v1 kind: HelmRepository metadata: name: metrics-server namespace: kube-system spec: interval: 1h url: https://kubernetes-sigs.github.io/metrics-server/ --- apiVersion: helm.toolkit.fluxcd.io/v2 kind: HelmRelease metadata: name: metrics-server namespace: kube-system spec: interval: 30m chart: spec: chart: metrics-server version: "3.x" sourceRef: kind: HelmRepository name: metrics-server values: args: - --kubelet-insecure-tls
3. Helm from OCI Registry (Recommended)
apiVersion: source.toolkit.fluxcd.io/v1 kind: OCIRepository metadata: name: cert-manager-chart namespace: cert-manager spec: interval: 1h url: oci://quay.io/jetstack/charts/cert-manager layerSelector: mediaType: "application/vnd.cncf.helm.chart.content.v1.tar+gzip" operation: copy ref: semver: "1.x" --- apiVersion: helm.toolkit.fluxcd.io/v2 kind: HelmRelease metadata: name: cert-manager namespace: cert-manager spec: interval: 1h chartRef: kind: OCIRepository name: cert-manager-chart install: strategy: name: RetryOnFailure retryInterval: 5m upgrade: strategy: name: RetryOnFailure retryInterval: 5m values: crds: enabled: true
4. FluxInstance with OCI Sync (Gitless GitOps)
apiVersion: fluxcd.controlplane.io/v1 kind: FluxInstance metadata: name: flux namespace: flux-system spec: distribution: version: "2.x" registry: "ghcr.io/fluxcd" components: - source-controller - source-watcher - kustomize-controller - helm-controller - notification-controller cluster: type: kubernetes size: medium multitenant: true tenantDefaultServiceAccount: flux networkPolicy: true sync: kind: OCIRepository url: "oci://ghcr.io/my-org/fleet-manifests" ref: "latest" path: "clusters/production" pullSecret: "registry-auth"
5. ResourceSet for Multi-Component Orchestration
apiVersion: fluxcd.controlplane.io/v1 kind: ResourceSet metadata: name: apps namespace: flux-system annotations: fluxcd.controlplane.io/reconcileEvery: "5m" spec: dependsOn: - apiVersion: fluxcd.controlplane.io/v1 kind: ResourceSet name: infra ready: true inputs: - tenant: "frontend" tag: "latest" environment: "production" - tenant: "backend" tag: "latest" environment: "production" resources: - apiVersion: v1 kind: Namespace metadata: name: << inputs.tenant >> labels: toolkit.fluxcd.io/role: "tenant" - apiVersion: v1 kind: ServiceAccount metadata: name: flux namespace: << inputs.tenant >> - apiVersion: rbac.authorization.k8s.io/v1 kind: RoleBinding metadata: name: flux namespace: << inputs.tenant >> roleRef: apiGroup: rbac.authorization.k8s.io kind: ClusterRole name: admin subjects: - kind: ServiceAccount name: flux namespace: << inputs.tenant >> - apiVersion: source.toolkit.fluxcd.io/v1 kind: OCIRepository metadata: name: apps namespace: << inputs.tenant >> spec: interval: 5m url: "oci://ghcr.io/my-org/apps/<< inputs.tenant >>" ref: tag: << inputs.tag >> - apiVersion: kustomize.toolkit.fluxcd.io/v1 kind: Kustomization metadata: name: apps namespace: << inputs.tenant >> spec: targetNamespace: << inputs.tenant >> serviceAccountName: flux interval: 30m retryInterval: 5m wait: true timeout: 5m sourceRef: kind: OCIRepository name: apps path: "./<< inputs.environment >>" prune: true
6. Image Automation
Flux supports two delivery models for updating container images and Helm chart versions. Pick based on whether the team wants Git commits as the audit log for version changes:
- Git-based —
+ImageRepository
+ImagePolicy
scan the registry and commit tag bumps back to Git viaImageUpdateAutomation
YAML markers. Requires$imagepolicy
andimage-reflector-controller
on the cluster. Loadimage-automation-controller
.references/image-automation.md - Gitless —
+ResourceSet
(ResourceSetInputProvider
) scans the registry and re-renders thetype: OCIArtifactTag
directly, upgrading the downstreamResourceSet
orHelmRelease
without touching Git. No bot credentials, no Git poll lag, no extra controllers. Recommended default for Flux Operator deployments. LoadKustomization
.references/gitless-image-automation.md
Gitless is the better fit when the tag lives in Helm values, when tags should differ per cluster in a fleet, or when the team doesn't want a bot writing to the repo. Git-based is the better fit when PR-based approval of version bumps is required or when Git must remain the canonical record of every deployed version.
7. Notifications (Slack, GitHub, Webhooks)
Provider + Alert for outgoing notifications, Receiver for incoming webhooks. Alert and Provider use
v1beta3, Receiver uses v1.
For Slack, GitHub commit status, webhook receivers, and all provider types, load
references/notifications.md.
Common Mistakes
Wrong template delimiters:
- ResourceSet uses
— NOT<< inputs.field >>
or{{ .inputs.field }}{{ inputs.field }} - Go templates
are only used in ImageUpdateAutomation{{ }}.spec.git.commit.messageTemplate
Mutual exclusivity:
- HelmRelease:
andspec.chart.spec
are mutually exclusivespec.chartRef - FluxInstance: only one per cluster, must be named
flux
HelmRelease strategy fields:
- Install/upgrade strategy is at
andspec.install.strategy.namespec.upgrade.strategy.name - Always use
— it retries without rollback or uninstall, avoiding downtimeRetryOnFailure - Do not use
orRemediateOnFailure
/spec.install.remediationspec.upgrade.remediation
OCIRepository for Helm charts:
- When using OCIRepository to fetch Helm charts from OCI registries, set
to extract the chart:layerSelectorlayerSelector: mediaType: "application/vnd.cncf.helm.chart.content.v1.tar+gzip" operation: copy
Reference Index
Load reference files and OpenAPI schemas based on the question topic. Load at most 1-2 reference files per question. Read schemas for field-level validation when generating YAML.
| CRD | Reference | Schema |
|---|---|---|
| FluxInstance | | |
| FluxReport | | |
| ResourceSet | | |
| ResourceSetInputProvider | | |
| GitRepository | | |
| OCIRepository | | |
| HelmRepository | | |
| HelmChart | | |
| Bucket | | |
| ExternalArtifact | | |
| ArtifactGenerator | | |
| Kustomization | | |
| HelmRelease | | |
| Provider | | |
| Alert | | |
| Receiver | | |
| ImageRepository | | |
| ImagePolicy | | |
| ImageUpdateAutomation | | |
| Topic | Reference |
|---|---|
| Repository structure, monorepo vs multi-repo, OCI-based fleet management | |
| Best practices, dependency management, remediation, versioning | |
| Web UI, dashboard, SSO, OIDC, Dex, Keycloak, Entra ID, RBAC | |
| MCP Server, AI assistant integration, in-cluster deployment | |
| Terraform bootstrap of Flux Operator | |
| Gitless image automation (ResourceSet + OCIArtifactTag) | |