Helm Chart Generator

Overview

Generate production-ready Helm charts with best practices built-in. Create complete charts or individual resources with standard helpers, proper templating, and automatic validation.

Official Documentation:

• Helm Docs | Chart Best Practices | Template Functions | Sprig Functions

Philosophy

• Ask rather than assume: critical settings such as image, ports, limits and probes come from the user.
• Build on helpers: names and labels come from _helpers.tpl, never from repeated literals.
• Keep the chart environment-neutral: defaults in values.yaml, environment overrides outside the chart.
• Validate rendered output, not just the templates.

When to Use

• Creating or scaffolding a new Helm chart with Chart.yaml, values.yaml and templates.
• Writing individual templates such as Deployments, Services, Ingress, HPAs or .tpl helpers.
• Converting raw Kubernetes manifests into a parameterised chart.
• Adding custom resources (ServiceMonitor, Certificate and similar) to a chart.

When Not to Use

• Validating or linting an existing chart: use devops-skills:helm-validator instead.
• Writing raw Kubernetes YAML with no Helm: use k8s-generator instead.
• Debugging a failed release or cluster state.
• Writing Terraform, Kustomize or other deployment tooling.

Chart Generation Workflow

Stage 1: Understand Requirements

REQUIRED: Use AskUserQuestion if any of these are missing or ambiguous:

Do NOT assume values for critical settings. Ask first, then proceed.

Stage 2: CRD Documentation Lookup

If custom resources are needed:

1. Try context7 MCP first:

   mcp__context7__resolve-library-id with operator name
   mcp__context7__get-library-docs with topic for CRD kind

2. Fallback to WebSearch:

   "<operator>" "<CRD-kind>" "<version>" kubernetes documentation spec

See references/crd_patterns.md for common CRD examples.

Stage 3: Create Chart Structure

Use the scaffolding script:

bash scripts/generate_chart_structure.sh <chart-name> <output-directory> [options]

Script options:

• --image <repo> - Image repository (default: nginx). Note: Pass only the repository name without tag (e.g., redis not redis:7-alpine)
• --port <number> - Service port (default: 80)
• --type <type> - Workload type: deployment, statefulset, daemonset (default: deployment)
• --with-templates - Generate resource templates (deployment.yaml, service.yaml, etc.)
• --with-ingress - Include ingress template
• --with-hpa - Include HPA template
• --force - Overwrite existing chart without prompting

Important customization notes:

• The script uses http as the default port name in templates. Customize port names for non-HTTP services (e.g., redis, mysql, grpc)
• Templates include checksum annotations for ConfigMap/Secret changes (conditionally enabled via .Values.configMap.enabled and .Values.secret.enabled)

Stage 4: Generate Standard Helpers

Use the helpers script or assets/_helpers-template.tpl:

bash scripts/generate_standard_helpers.sh <chart-name> <chart-directory>

Stage 5: Generate Templates

⚠️ CRITICAL REQUIREMENT: Read Reference Files NOW

You MUST use the Read tool to load these reference files at this stage, even if you read them earlier in the conversation:

1. Read references/resource_templates.md - for the specific resource type patterns
2. Read references/helm_template_functions.md - for template function usage
3. Read references/crd_patterns.md - if generating CRD resources (ServiceMonitor, Certificate, etc.)

Why: Prior context may be incomplete or summarized. Reading reference files at generation time guarantees all patterns, functions, and examples are available for accurate template creation.

Do NOT skip this step. Template quality depends on having current reference patterns loaded.

Reference templates for all resource types in references/resource_templates.md:

• Workloads: Deployment, StatefulSet, DaemonSet, Job, CronJob
• Services: Service, Ingress
• Config: ConfigMap, Secret
• RBAC: ServiceAccount, Role, RoleBinding, ClusterRole, ClusterRoleBinding
• Network: NetworkPolicy
• Autoscaling: HPA, PodDisruptionBudget

Key patterns (MUST include in all templates):

# Use helpers for names and labels
metadata:
  name: {{ include "mychart.fullname" . }}
  labels: {{- include "mychart.labels" . | nindent 4 }}

# Conditional sections with 'with'
{{- with .Values.nodeSelector }}
nodeSelector: {{- toYaml . | nindent 2 }}
{{- end }}

# Checksum annotation (REQUIRED for Deployments/StatefulSets/DaemonSets to trigger restarts on config changes)
annotations:
  checksum/config: {{ include (print $.Template.BasePath "/configmap.yaml") . | sha256sum }}

Stage 6: Create values.yaml

Structure guidelines:

• Group related settings logically
• Document every value with # -- comments
• Provide sensible defaults
• Include security contexts, resource limits, probes

See assets/values-schema-template.json for JSON Schema validation.

Stage 7: Validate

Run validation using devops-skills:helm-validator skill (helm lint, template render, schema checks, dry-run).

Template Functions Quick Reference

See references/helm_template_functions.md for complete guide.

Working with CRDs

See references/crd_patterns.md for complete examples. Ship CRDs in crds/ directory (not templated); template CR instances in templates/.

Converting Manifests to Helm

1. Parameterize names (use helpers) and extract values
2. Apply label/conditional patterns, use toYaml for complex objects
3. Create _helpers.tpl with standard helpers
4. Validate with devops-skills:helm-validator

Error Handling

Post-Generation Validation

After generating charts, invoke devops-skills:helm-validator to ensure quality.

Anti-Patterns

NEVER hardcode image tags in values.yaml

WHY: Pinning to :latest or a hard-coded version in the chart prevents version overrides at deploy time.

BAD: image: repository: myapp tag: latest

GOOD: image: repository: myapp tag: "" with appVersion as the default, overridden via --set image.tag=v1.2.3.

NEVER omit resources: limits and requests on containers

WHY: Containers without resource constraints are evicted unpredictably under node pressure and cannot be scheduled by the Kubernetes cluster autoscaler.

BAD: No resources: block in the container spec template.

GOOD: Set both requests and limits for CPU and memory, with documented tuning guidance in values.yaml.

NEVER use helm upgrade --install without --atomic in CI/CD

WHY: Without --atomic, a failed upgrade leaves the release in a broken state that blocks future upgrades and requires manual helm rollback.

BAD: helm upgrade --install myapp ./chart

GOOD: helm upgrade --install --atomic --timeout 5m myapp ./chart

NEVER place environment-specific values inside the chart's default values.yaml

WHY: Mixing production values into the chart couples the chart to one environment.

BAD: Production database URLs in values.yaml.

GOOD: Use a layered values approach: values.yaml for defaults, values-prod.yaml for overrides, -f values-prod.yaml at deploy time.

NEVER skip helm template + kubeval/kubeconform validation

WHY: A chart that renders without error can still produce invalid Kubernetes manifests.

BAD: Only run helm lint before deploying.

GOOD: helm template . | kubeval --strict --ignore-missing-schemas to validate rendered manifests against the Kubernetes API schema.

NEVER assume values for critical settings

WHY: guessed images, ports, limits or probe paths produce a chart that installs but does not work.

BAD: silently choosing nginx and port 80 when the user did not say.

GOOD: ask for the missing image, port, limits, probes, scaling, workload type and storage before generating.

NEVER skip reading the reference files at the template stage

WHY: earlier context may be incomplete or summarised, so patterns and function usage drift from the references.

BAD: writing templates from memory because the references were read earlier.

GOOD: read the resource templates, template functions and (for CRDs) CRD patterns at generation time.

NEVER hardcode resource names or labels in templates

WHY: literal names and duplicated labels drift between resources and break release naming.

BAD: name: myapp and hand-written app: myapp labels.

GOOD: {{ include "mychart.fullname" . }} and {{- include "mychart.labels" . | nindent 4 }}.

NEVER omit the checksum annotation on workloads that consume config

WHY: without it, Deployments, StatefulSets and DaemonSets do not restart when a ConfigMap or Secret changes.

BAD: a Deployment that mounts a ConfigMap with no checksum/config annotation.

GOOD: add the checksum annotation, gated by .Values.configMap.enabled and .Values.secret.enabled.

NEVER put CRDs in templates/

WHY: CRDs must exist before their instances and should not be templated or re-rendered per release.

BAD: a CRD manifest under templates/.

GOOD: ship CRDs in the crds/ directory and template only the custom resource instances.

NEVER leave the default http port name on a non-HTTP service

WHY: the scaffolding script names the port http, which misleads probes, services and operators for other protocols.

BAD: a Redis service port named http.

GOOD: rename the port after the protocol, for example redis, mysql or grpc.

References