Frontend Forge Fe Operations Agent Skill

Frontend Forge FE Operations

When to use

Use this skill for operational work around FrontendExtension resources:

• Create or update an FE manifest
• Trigger or inspect package generation
• Force a rebuild with the rebuild-token annotation
• Download the generated package artifact
• Publish, unpublish, or delete an extension
• Inspect FE status, conditions, labels, Jobs, artifact ConfigMaps, and publish targets
• Debug FE package phases such as Packaging, Ready, and Failed, plus publish/unpublish phases such as Pending, Running, Succeeded, and Failed

If the task is about FrontendIntegration runtime JSBundle creation, use the FI operations skill instead. FE package/publish does not create a runtime JSBundle in the current cluster.

Quick Entry

• Create or update FE -> read references/lifecycle.md
• Rebuild package -> read references/lifecycle.md
• Publish, unpublish, delete-with-unpublish, list through API, or download -> read references/api.md
• Inspect package Job, artifact ConfigMap, publish Job, labels, or conditions -> read references/inspection.md
• Debug stuck or failed state -> read references/inspection.md, then check controller logs

Preconditions

• The current Kubernetes context is pointed at the target cluster.
• The FE CRD exists: frontendextensions.frontend-forge.kubesphere.io.
• The FE controller is installed, usually as the extension-controller component.
• The FE API is installed when using download, publish, unpublish, or delete-with-unpublish HTTP operations.
• The build service configured by BUILD_SERVICE_BASE_URL is reachable from package Jobs.
• Publish target data exists when publishing, normally ConfigMap/ksbuilder-publish-config in extension-frontend-forge.

Source of truth

Prefer live cluster state for operations and repo docs for expected behavior:

• kubectl get fe <name> -o yaml
• kubectl get jobs -n <work-namespace> -l frontend-forge.kubesphere.io/fe-name=<name>
• spec/frontend-extension-design.md
• spec/crds.md
• spec/k8s-resources.md
• config/samples/frontendextension-inspecttask.yaml
• crates/api/src/fe.rs
• crates/frontend-extension-controller
• crates/frontend-forge-extension-api

Resource Model

• FrontendExtension
• cluster-scoped
• short name: fe
• source object for package, artifact, download, publish, and unpublish state
• Package Job
• namespaced
• default namespace: extension-frontend-forge
• typical name: fe-<fe-name>-package-<artifact-key-12>-a<attempt>
• Artifact ConfigMap
• namespaced
• default namespace: extension-frontend-forge
• referenced by status.artifact.storage.ref
• contains binaryData["package.tgz"], data["artifact.json"], and data["files.json"]
• Publish or unpublish Job
• namespaced
• default namespace: extension-frontend-forge
• typical names:
• fe-<fe-name>-publish-<request-id-hash-short>
• fe-<fe-name>-unpublish-<request-id-hash-short>
• Publish target
• ConfigMap or Secret
• default chart target is usually ConfigMap/ksbuilder-publish-config in the release namespace

Default names and namespaces can change through Helm values and controller environment variables. Use FE status, Job labels, and controller deployment env vars before assuming defaults.

Common Commands

Inspect FE:

kubectl get fe <name> -o yaml
kubectl get fe <name> -o jsonpath='{.status}{"\n"}'
kubectl get fe <name> -o jsonpath='{.status.phase}{" "}{.status.publish.phase}{" "}{.status.unpublish.phase}{"\n"}'
kubectl get fe -l frontend-forge.kubesphere.io/package-state=ready
kubectl get fe -l frontend-forge.kubesphere.io/publish-state=published

Create or update:

kubectl apply -f <file.yaml>
kubectl apply -f config/samples/frontendextension-inspecttask.yaml

Find package Job and logs:

kubectl get fe <name> -o jsonpath='{.status.packageJob.namespace}{" "}{.status.packageJob.name}{"\n"}'
kubectl -n <job-namespace> get job <job-name> -o yaml
kubectl -n <job-namespace> logs job/<job-name>

Find artifact ConfigMap:

kubectl get fe <name> -o jsonpath='{.status.artifact.storage.ref.namespace}{" "}{.status.artifact.storage.ref.name}{" "}{.status.artifact.storage.key}{"\n"}'
kubectl -n <artifact-namespace> get cm <artifact-configmap-name> -o yaml

Force a rebuild:

kubectl annotate fe <name> frontend-forge.kubesphere.io/rebuild-token="$(date +%s)" --overwrite

Inspect publish or unpublish Jobs:

kubectl get jobs -n extension-frontend-forge -l frontend-forge.kubesphere.io/fe-name=<name>
kubectl get fe <name> -o jsonpath='{.status.publish.jobRef.namespace}{" "}{.status.publish.jobRef.name}{"\n"}'
kubectl get fe <name> -o jsonpath='{.status.unpublish.jobRef.namespace}{" "}{.status.unpublish.jobRef.name}{"\n"}'
kubectl -n <job-namespace> logs job/<job-name>

FE API operations:

KS_API=https://<kubesphere-host>
FE_API="$KS_API/kapis/frontend-forge-api.kubesphere.io/v1alpha1/frontendextensions"
curl -fS "$FE_API/<name>"
curl -fS -u "user:password" "$FE_API/<name>"  # user runs this if /kapis returns 401/403
curl -fS "$FE_API/<name>/publish"      # read publish status only
curl -fS -X POST -H 'Content-Type: application/json' --data '{"requestId":"manual-1","expectedArtifactDigest":"sha256:<digest>"}' "$FE_API/<name>/publish"
curl -fS "$FE_API/<name>/unpublish"    # read unpublish status only
curl -fS -X POST -H 'Content-Type: application/json' --data '{"requestId":"manual-unpublish-1"}' "$FE_API/<name>/unpublish"
curl -fS -X POST -H 'Content-Type: application/json' --data '{"unpublish":true}' "$FE_API/<name>/delete"
curl -fS -u "user:password" -X POST -H 'Content-Type: application/json' --data '{"unpublish":true}' "$FE_API/<name>/delete"
curl -fL "$FE_API/<name>/download" -o <name>.tgz

Operating Rules

• Do not use -n with kubectl get fe; FrontendExtension is cluster-scoped.
• Use the FE HTTP API for publish, unpublish, download, and delete-with-unpublish when available. It validates artifact readiness, publish target, digest expectations, and idempotency.
• For KubeSphere users, prefer /kapis/frontend-forge-api.kubesphere.io/v1alpha1/frontendextensions; use direct service port-forwarding mainly for local debugging.
• If /kapis returns 401 or 403, guide the user to run curl -u "user:password" ... or use their normal KubeSphere authenticated session. Do not ask them to paste credentials into the conversation.
• Treat publish and unpublish annotations as implementation details written by the FE API. Inspect them for debugging, but do not make raw annotation patches the default workflow.
• Directly patch publish/unpublish annotations only as a last-resort recovery step when the API is unavailable and the operator explicitly accepts the risk. Missing or stale generation, source hash, artifact digest, or target annotations can produce failed publish status.
• Do not expect package creation to publish the extension. Packaging and publishing are separate flows.
• Do not expect direct kubectl delete fe <name> to unpublish first. Use the FE API delete endpoint with {"unpublish":true} for that behavior.
• Treat status.artifact.artifactKey as build cache identity, not package content digest. Use status.artifact.digest for the generated package digest.
• Treat artifact ConfigMaps as controller-owned outputs. Read them for inspection, but do not edit or delete the ConfigMap referenced by status.artifact.storage.ref unless the user has explicitly requested artifact cleanup and you have confirmed it is not the current artifact.

Safety Rules

Destructive operations:

• Before deleting an FE, check .status.publish.phase, .status.publish.active, and .status.publish.artifactDigest.
• If a published extension should be removed, use the FE API delete endpoint with {"unpublish":true} so the controller can unpublish before deleting.
• Avoid kubectl delete fe <name> for published extensions unless the user explicitly wants to skip unpublish.
• Before deleting Jobs or artifact ConfigMaps, verify owner refs, labels, and whether FE status still references them.

Rebuild:

• Rebuild changes artifact cache identity and can make a previously published artifact stale. Before setting frontend-forge.kubesphere.io/rebuild-token, record current .status.observedSourceHash, .status.artifact.digest, .status.artifact.artifactKey, .status.publish, and frontend-forge.kubesphere.io/publish-fresh.
• Prefer a unique rebuild token, such as a timestamp or incident id, and use --overwrite.
• After rebuild, verify the new package reaches Ready; republish only after confirming the new status.artifact.digest.

Artifact ConfigMap:

• Use the FE API download endpoint for package bytes when possible; it checks phase, download readiness, source hash, storage kind, and digest.
• Do not mutate binaryData["package.tgz"], data["artifact.json"], or data["files.json"] in place. A manual edit can make status, annotations, and digest disagree.
• Do not delete the ConfigMap named by .status.artifact.storage.ref during normal troubleshooting. If cleanup is requested, first confirm the current FE artifact points somewhere else or the FE itself is being deleted.

Status Fields To Check

FE package state:

• .status.phase
• .status.observedGeneration
• .status.observedSourceHash
• .status.observedRebuildToken
• .status.conditions
• .status.packageJob
• .status.artifact
• .status.download

Publish state:

• .status.publish.phase
• .status.publish.active
• .status.publish.requestId
• .status.publish.artifactDigest
• .status.publish.jobRef
• .status.publish.lastError
• .status.unpublish.phase
• .status.unpublish.requestId
• .status.unpublish.extensionName
• .status.unpublish.jobRef
• .status.unpublish.lastError

Labels used for filtering:

• frontend-forge.kubesphere.io/package-state: packaging, ready, failed
• frontend-forge.kubesphere.io/publish-state: not-published, publishing, published, failed
• frontend-forge.kubesphere.io/publish-fresh: true, false

Status consistency with implementation:

• FE package phase is only Pending, Packaging, Ready, or Failed.
• Package Job phase is Pending, Running, Succeeded, or Failed.
• Publish phase is NotRequested, Pending, Running, Succeeded, or Failed.
• Unpublish phase is NotRequested, Pending, Running, Succeeded, or Failed.
• Controller condition types are SourceValid, ArtifactReady, DownloadReady, and PublishSucceeded.
• PublishFailed appears as a PublishSucceeded condition reason when status.publish.phase=Failed; it is not a top-level FE phase.

Troubleshooting Order

• Inspect the FE object and status first.
• If package state is not Ready, inspect status.conditions, status.packageJob, package Job logs, build-service reachability, and artifact ConfigMap state.
• If package state is Ready but download fails, inspect status.download, status.artifact.storage, the artifact ConfigMap key, and digest consistency.
• If publish or unpublish fails, inspect status.publish or status.unpublish, publisher Job logs, target ConfigMap or Secret contents, and API responses. Use annotations as diagnostic evidence, not as the primary operation path.
• If live behavior differs from expected defaults, inspect controller deployment env vars and Helm values before changing the FE.

Controller Logs

When status and Job logs are insufficient:

kubectl -n extension-frontend-forge logs deploy/frontend-forge-extension-controller --tail=200
kubectl -n extension-frontend-forge logs deploy/frontend-forge-extension-api --tail=200

Deployment names can vary by Helm release name. If those commands fail, list deployments with:

kubectl -n extension-frontend-forge get deploy -l app.kubernetes.io/component=extension-controller
kubectl -n extension-frontend-forge get deploy -l app.kubernetes.io/component=extension-api