Convert Docker Compose files or installation docs into production-grade Sealos templates with role-specific personal low-load resource sizing, official route and runtime semantics, KubeBlocks database and Job gates, and managed or optional S3 storage contracts. Use when user has a docker-compose.yml and wants a Sealos or Kubernetes template, wants to migrate from Docker Compose to Sealos, needs to convert container orchestration configs to Sealos format, or mentions compose-to-template conversion. Also triggers on "/docker-to-sealos".
Convert Docker Compose files or installation docs into production-grade Sealos templates with role-specific personal low-load resource sizing, official route and runtime semantics, KubeBlocks database and Job gates, and managed or optional S3 storage contracts. Use when user has a docker-compose.yml and wants a Sealos or Kubernetes template, wants to migrate from Docker Compose to Sealos, needs to convert container orchestration configs to Sealos format, or mentions compose-to-template conversion. Also triggers on "/docker-to-sealos".
Docker to Sealos Template Converter
Overview
Convert Docker Compose files or installation docs into production-grade Sealos templates.
Execute end-to-end automatically (analysis, conversion, validation, output) without asking users for missing fields.
Governance and Rule Priority
Use the following precedence to prevent rule drift:
SKILL.md MUST rules (this file)
references/sealos-specs.md and references/database-templates.md
references/conversion-mappings.md and references/example-guide.md
If lower-priority references conflict with higher-priority MUST rules, update the lower-priority files.
Do not keep conflicting examples.
Workflow
Step 1: Analyze input
Extract from Docker Compose/docs:
application services vs database services
volumes/config mounts/object storage requirements
ports, dependencies, service communication
env vars and secret usage
startup-time validation rules for bootstrap credentials, API keys, salts, secrets, and feature flags
account bootstrap mode from the exact selected release: functional first-user signup, mandatory bootstrap credentials, or optional root reconciliation/bootstrap
multi-service web roles: browser entry, REST API, OpenAI/API gateway, docs, workers, and one-shot jobs
resource limits/requests and health checks
if official Kubernetes installation docs/manifests are available, also extract app-runtime behavior from them (bootstrap admin fields, external endpoint/protocol assumptions, health probes, startup/init flow, migration ordering)
if official compose/docs provide multiple cooperating services, record the official runtime bundle source, component list, image versions, public entry routes, and critical env vars
record the selected source topology: topology-bearing resource roles, feature conditions, and application or database component replica counts
Step 2: Infer metadata
Infer and normalize:
app name, title, description, categories
official URL, gitRepo, icon source (prefer square/circular icon-first assets such as app icons, favicons, or avatars; avoid rectangular wordmark/text logos)
locale/i18n metadata
Step 3: Plan resources in strict order
Generate resources in this order:
Template CR
ObjectStorageBucket (if needed)
Database resources (ServiceAccount → Role → RoleBinding → Cluster → Job if needed)
App workload resources (ConfigMap/Secret → Deployment/StatefulSet → Service → Ingress)
App resource (last)
Step 4: Apply conversion rules
Apply field-level mappings from references/conversion-mappings.md, including:
image pinning and annotation mapping
port/service/ingress conversion
env var conversion and dependency ordering
storage conversion and vn naming (scripts/path_converter.py)
service-name to Kubernetes FQDN conversion
for DB URL/DSN envs (for example *_DATABASE_URL, *_DB_URL), when Kubeblocks endpoint is host:port, inject host/port/username/password via approved secretKeyRef envs and compose the final URL with $(VAR) expansion
edge gateway normalization: when Compose includes Traefik-like edge proxy plus business services, skip the proxy workload and expose business services via Sealos Ingress directly
TLS offload normalization for Sealos Ingress: when a business service exposes both 80 and 443, drop 443 from workload/service ports and remove in-container TLS certificate mounts (for example /etc/nginx/ssl, /etc/ssl, /certs) unless official Kubernetes docs explicitly require HTTPS backend-to-service traffic
multi-service web normalization: expose the verified browser entry in the App resource, expose API/gateway/docs only when they are intended public surfaces, and keep workers private with no Service/Ingress
URL topology: browser-facing env vars must use public HTTPS URLs, while server-to-server env vars must use Kubernetes Service FQDNs unless the app explicitly requires public callbacks
WebSocket ingress normalization: when the public entry is ws://, wss://, CDP/Chrome DevTools, a game socket, or a WebSocket-named port/service, expose it with WebSocket nginx ingress annotations
StatefulSet service identity: for a single-component app with no documented headless or stable per-Pod DNS requirement, use the public application Service as spec.serviceName and keep the workload, Service, root Ingress, and manager identity aligned; preserve documented HA/headless governing Services and expose them through a separate public application Service
prefer scripts/compose_to_template.py --kompose-mode always as deterministic conversion entrypoint (require kompose for reproducible workload shaping)
for existing-template updates, keep the current template's topology-bearing resources, feature conditions, and replica counts as the baseline
for new conversions, keep the selected Compose services and deploy.replicas values as the topology baseline
use official Kubernetes installation docs/manifests to align app-runtime semantics such as bootstrap fields, endpoints, probes, and startup ordering
keep optional or recommended workers, caches, and HA replicas outside the emitted topology unless the selected source topology or explicit user intent includes them
keep every feature input scoped to its documented capability; database and object-storage inputs must not add unrelated workloads, caches, or replicas
when official compose/docs define a multi-component runtime bundle, keep runtime-required components, entry routes, critical env vars, and component image versions aligned to one official release/compose source
before converting a host directory mount to persistent storage, verify whether the image already ships required files at that target path; avoid hiding image-bundled manifests, dependency lists, or config defaults behind a fresh empty PVC
Step 5: Apply database strategy
Database services must be generated as KubeBlocks Cluster resources. Do not convert PostgreSQL/MySQL/MongoDB/Redis/Kafka Compose database services into raw Kubernetes Deployment or StatefulSet workloads.
PostgreSQL must follow the pinned version and structure requirements.
MySQL/MongoDB/Redis/Kafka must use templates and secret naming from references/database-templates.md.
Add DB init Job/initContainer when application database bootstrap requires it.
For PostgreSQL custom databases (non-postgres), the init Job must wait for PostgreSQL readiness before execution and create the target database idempotently.
Database client images may be used in app initContainers and init/migration/bootstrap Jobs for readiness and bootstrap gates.
Critical application compatibility objects must be verified in live database state. Use idempotent initContainer self-healing for compatibility views, legacy tables/views, indexes, extensions, search paths, and bootstrap state that the app requires on every cold start.
One-shot init Jobs may create initial databases or seed state, but app startup gates must verify the final database objects directly. Treat TTL-expired Jobs as historical evidence and rely on database state for acceptance.
Worker, gateway, and background services that depend on app migrations must wait for the required tables, migration markers, or app-specific readiness objects, not only for the database port.
Redis readiness probes or initContainers must tolerate authenticated Redis responses such as NOAUTH or Authentication required when credentials are not needed for readiness.
PostgreSQL bootstrap shell must use safe quoting patterns. Prefer shell-level existence checks plus simple SQL statements when possible. Use single-quoted heredocs or SQL files for psql variable interpolation, and avoid PL/pgSQL DO $$ blocks in inline shell commands when a guard query can express the same logic.
Do not use psql -c "..." for :'var' variable interpolation. Use psql -v name=value <<'SQL' ... :'name' ... SQL or pass already-safe literal SQL.
Step 6: Generate output files
Always produce:
template/<app-name>/index.yaml
template/<app-name>/logo.<ext> when official icon is resolvable, prioritizing square/circular icon-first artwork and avoiding rectangular wordmark/text logos
Never create:
template/<app-name>/README.md
template/<app-name>/README_zh.md
README authoring is out of scope for this skill. If the Template CR requires README URLs, populate URL fields in index.yaml only and leave file creation to a dedicated README skill.
Step 7: Validate before output
Run validator and self-tests before delivering template output.
If validation fails, fix template/rules/examples first.
For web applications, live validation must include runtime log hygiene: inspect init and main container logs after first readiness, after login or setup, and after one documented API negative route or unique missing-static-asset request. SPA client routes may return the HTML shell with HTTP 200; recurring traceback-style warnings remain template failures even when Pods are Ready.
For login-gated web applications, live validation must prove the real credential/session flow with one authenticated API or page before resource tuning or cleanup.
For managed or private object storage, live validation must upload known bytes through the authenticated application flow, read or download the object, compare its SHA-256 digest, confirm delivery through the application proxy or a time-bounded presigned URL, and verify the raw anonymous object request remains restricted. Optional object storage must validate the local-storage and managed-bucket branches independently.
MUST Rules (Condensed)
Naming and metadata
Template metadata.name must be hardcoded lowercase; do not use ${{ defaults.app_name }}.
Template CR folder name must match metadata.name.
Template CR must include required metadata fields (title, url, gitRepo, author, description, icon, templateType, locale, i18n, categories).
Template spec.readme must point to https://raw.githubusercontent.com/labring-actions/templates/kb-0.9/template/<app-name>/README.md.
Template spec.i18n.zh.readme must point to https://raw.githubusercontent.com/labring-actions/templates/kb-0.9/template/<app-name>/README_zh.md.
These README fields are URL references in index.yaml only; this skill must not create or update the referenced README files.
icon URL must point to template repo raw path for this app on kb-0.9 branch.
template/<app-name>/logo.<ext> must use square/circular icon-first artwork (for example app icon/favicon/avatar), and must not use rectangular wordmark/text logos.
i18n.zh.description must be written in Simplified Chinese.
Omit i18n.zh.title when it is identical to title.
categories must only use predefined values (tool, ai, game, database, low-code, monitor, dev-ops, blog, storage, frontend, backend).
App resource
App resource must use spec.data.url.
App resource spec.displayType must be normal.
App resource spec.type must be link.
App resource spec.data.url must be the browser entry URL that succeeds from a fresh Sealos launch. For apps with safe-path, setup-path, or entrance-path behavior, verify the configured path and root path, then choose the URL that supports login or first-run setup without hidden prior navigation.
SSR/Next.js/React server apps must not use a path that renders a server-side exception as the App URL or HTTP probe. Treat visible Application error, server-side exception, Internal Server Error, or Unhandled Runtime Error text as a failed entry path even if the HTTP status is 2xx/3xx.
Never use spec.template in App resource.
cloud.sealos.io/app-deploy-manager label value must equal resource metadata.name.
metadata.labels.app label value must equal resource metadata.name for managed app workloads.
The primary business container name must equal workload metadata.name for managed app workloads; sidecar/helper containers may use distinct descriptive names.
Application Service resources must define metadata.labels.app and metadata.labels.cloud.sealos.io/app-deploy-manager, and both labels must match spec.selector.app.
Runtime component-scoped ConfigMap resources must define metadata.labels.app and metadata.labels.cloud.sealos.io/app-deploy-manager, and both labels must match metadata.name; bootstrap-only ConfigMaps used only by init containers to copy initial config into persistent storage must not define either label.
Application Service resources must use the same component name across metadata.name, metadata.labels.app, metadata.labels.cloud.sealos.io/app-deploy-manager, and spec.selector.app.
Root-path Ingress resources (pathType: Prefix, path: /) must use the same component name across metadata.name, metadata.labels.cloud.sealos.io/app-deploy-manager, and backend service.name; non-root or non-Prefix Ingress rules may route to a different backend service.
Root-path Ingress resources (pathType: Prefix, path: /) must use backend.service.port.number, and the number must match a declared spec.ports[*].port on the referenced application Service.
Root-path Prefix routes must be the first entry in each HTTP paths list so Launchpad public-address discovery selects the application entry route.
Service spec.ports[*].name must be explicitly set (required for multi-port services).
HTTP Ingress must include required nginx annotations (kubernetes.io/ingress.class, nginx.ingress.kubernetes.io/proxy-body-size, nginx.ingress.kubernetes.io/server-snippet, nginx.ingress.kubernetes.io/ssl-redirect, nginx.ingress.kubernetes.io/backend-protocol, nginx.ingress.kubernetes.io/client-body-buffer-size, nginx.ingress.kubernetes.io/proxy-buffer-size, nginx.ingress.kubernetes.io/proxy-send-timeout, nginx.ingress.kubernetes.io/proxy-read-timeout, nginx.ingress.kubernetes.io/configuration-snippet) with expected defaults.
WebSocket Ingress must include required nginx annotations (kubernetes.io/ingress.class, nginx.ingress.kubernetes.io/proxy-body-size, nginx.ingress.kubernetes.io/proxy-read-timeout, nginx.ingress.kubernetes.io/proxy-send-timeout, nginx.ingress.kubernetes.io/backend-protocol, nginx.ingress.kubernetes.io/ssl-redirect) with backend-protocol: WS and 3600 read/send timeouts.
CronJob resources must define labels cloud.sealos.io/cronjob, cronjob-launchpad-name, and cronjob-type; cloud.sealos.io/cronjob must equal metadata.name, cronjob-launchpad-name must be "", and cronjob-type must be image.
When official application health checks are available, managed workloads must define livenessProbe, readinessProbe, and (for slow bootstrap apps) startupProbe, aligned with official endpoints/commands.
For public images that are verified to run as a non-root UID, managed app workloads and init Jobs should set restricted-compatible security context (runAsNonRoot, runAsUser, runAsGroup, fsGroup, seccompProfile: RuntimeDefault, allowPrivilegeEscalation: false, capabilities.drop: [ALL]) unless the image requires root or extra capabilities.
Official Kubernetes alignment
If official Kubernetes installation docs/manifests are available, conversion must reference them and align critical runtime settings before emitting template artifacts.
When official Kubernetes docs/manifests and Compose differ, prefer official Kubernetes runtime semantics for app behavior (bootstrap admin fields, external endpoint/env/protocol, health probes), unless doing so violates higher-priority Sealos MUST/security constraints.
For existing-template updates, preserve the current template's topology-bearing resource inventory, conditions, and replica counts; for new conversions, preserve the selected Compose topology and deploy.replicas values.
Use official Kubernetes docs/manifests to align application runtime semantics; add optional or recommended workers, caches, and HA replicas only when the selected source topology or explicit user intent includes them.
Each application feature input must gate only resources and settings for that documented feature; database and object-storage inputs must not change unrelated workload inventory or replica counts.
Topology-sensitive validation must provide .sealos/topology-evidence/<app-name>.yaml as validator-only TopologyEvidence; final Sealos Template artifacts must stay free of topology validator metadata.
When official compose/docs provide a multi-component runtime bundle, template artifacts must preserve runtime-required components, public entry routes, critical env vars, and image versions from the same official release/compose source.
Templates using official multi-component runtime evidence must provide a separate RuntimeBundleEvidence YAML file during validation, while final Sealos Template artifacts stay free of runtime-bundle validator metadata.
Images and pull policy
Do not use :latest.
Resolve versions with crane: prefer an explicit version tag (for example v2.2.0), and fallback to digest pin only when a deterministic version tag is unavailable.
Avoid floating tags (for example :v2, :2.1, :stable); use an explicit version tag or digest.
Managed workload image references must be concrete and must not contain Compose-style variable expressions (for example ${VAR}, ${VAR:-default}); resolve to explicit tag or digest before emitting template artifacts.
Application originImageName must match container image.
Known public-image managed app workloads must omit template.spec.imagePullSecrets; when a registry-authenticated workload needs a pull Secret, it may reference only the app-scoped Secret ${{ defaults.app_name }}.
The registry pull Secret is runtime-managed by sealos-deploy using local gh CLI credentials for private GHCR images; do not expose raw registry credential inputs in generated templates.
All containers must explicitly set imagePullPolicy: IfNotPresent.
Storage
Do not use emptyDir.
Use persistent storage patterns (volumeClaimTemplates) where storage is needed.
StatefulSet resources with volumeClaimTemplates must keep standard workload labels such as app and cloud.sealos.io/app-deploy-manager, and omit only cloud.sealos.io/deploy-on-sealos from both StatefulSet metadata.labels and volumeClaimTemplates[].metadata.labels.
volumeClaimTemplates[].metadata must include a path-derived name, annotations.path, and annotations.value: '1', and each claim must match a container volumeMount with the same name and path.
PVC request must be <= 1Gi unless source spec explicitly requires less.
ConfigMap data keys must follow vn naming (scripts/path_converter.py), including /, -, ., and other special characters.
ConfigMaps mounted by managed Deployment/StatefulSet workloads must use metadata.name == workload.metadata.name.
ConfigMap workload volumes must use <workload-name>-cm, and every ConfigMap data key must be mounted as its own volumeMount with subPath exactly equal to that key.
Omit ConfigMap volume defaultMode in managed templates. Invoke mounted scripts through /bin/sh /path/script; copy to persistent storage and apply chmod in an initContainer when an application truly requires an executable file.
Avoid long inline startup scripts or heredocs in command/args; place initialization/start scripts in ConfigMap files and invoke them with a short command.
Classify object storage from official application docs before generating inputs: required capability, application-level optional capability, or externally managed storage.
If object storage/S3 integration is Enterprise, paid, commercial, subscription, or license-gated in the upstream application, keep the public template on the community-supported storage path (for example filesystem/PVC) and expose no standard ObjectStorageBucket or S3 input for that feature.
When object storage is required and Sealos S3 compatibility satisfies the application contract, create unconditional ObjectStorageBucket resources for the documented bucket topology and inject Sealos object-storage secrets.
A template with managed ObjectStorageBucket must use it as the sole object-store data plane and omit bundled MinIO server workloads, Services, Ingresses, PVCs, and local object-storage credentials.
Resolve object-storage provider/backend selector inputs during conversion, and do not combine a managed ObjectStorageBucket with bundled minio/minio, bitnami/minio, or bitnamilegacy/minio server images.
Use a compatibility proxy only when official protocol evidence requires request adaptation.
An object-storage compatibility proxy must declare metadata.annotations.docker-to-sealos.object-storage-compatibility-proxy-source as a credential-free HTTPS source URL or user-request:<reference>, remain stateless, and omit persistent volumes.
External S3/object-storage credential inputs require metadata.annotations.docker-to-sealos.external-object-storage-source as a credential-free HTTPS source URL or user-request:<reference>, and must not coexist with ObjectStorageBucket.
Managed or private object-storage acceptance must prove authenticated application upload and read/download with matching content, application-proxy or time-bounded presigned delivery, and restricted raw anonymous access; optional object storage must pass both local-storage and managed-bucket branches.
Env and secrets
Non-database sensitive values/inputs use direct env[].value.
When an official runtime profile constrains an env value's format or length, use a valid literal or a required input without a generated default; bare ${{ random(n) }} is invalid for hex- or encoding-constrained values.
Internal credentials that an official runtime library can deterministically derive from opaque entropy may use a quoted instance seed in spec.defaults; every consuming role must derive and validate the same final values, remove the seed before exec, and expose no user input for the derived values.
Persisted runtime-secret contracts marked with docker-to-sealos.runtime-secret-contract: persisted must generate from durable runtime entropy, apply restrictive permissions with an atomic replacement, validate before exec, and keep the final secret out of diagnostics.
When an official runtime profile selects an external provider, the workload must wire a non-empty required credential for that provider; an optional input with an empty default is invalid.
Business containers must source database connection fields (endpoint, host, port, username, password) from approved Kubeblocks database secrets via env[].valueFrom.secretKeyRef; exception: Redis host/port may use Sealos Redis Service FQDN and 6379 when the Redis secret only exposes credentials, and MongoDB host/port or connection URLs may use the Sealos MongoDB Service FQDN plus 27017 when the MongoDB secret exposes credentials only.
Business containers must not use custom env/volume Secret references except approved Kubeblocks database secrets and object storage secrets.
A dedicated app-scoped registry pull Secret is allowed only for private-registry images and must be referenced only through template.spec.imagePullSecrets; public images must not add pull secrets.
Database connection/bootstrap may use Kubeblocks-provided secrets, and reserved Kubeblocks database secret names must not be redefined by custom Secret resources.
Env vars must be declared before referenced (for example password before URL composition).
Follow official app env var naming; do not invent prefixes.
For split frontend/API/gateway apps, keep public browser URLs and internal service URLs separate. Frontend/browser callback variables use https://${{ defaults.<host> }}.${{ SEALOS_CLOUD_DOMAIN }}; backend-to-backend variables use http://<service>.$(SEALOS_NAMESPACE).svc.cluster.local:<port> or the fully rendered Service FQDN.
When the application requires its public URL configured via a file-based config system (e.g., node-config config/default.json, PHP config files), create a ConfigMap containing the config file with the public URL set to https://${{ defaults.app_host }}.${{ SEALOS_CLOUD_DOMAIN }}, and mount it to the application's config directory. The ConfigMap must follow standard naming and label conventions.
For PostgreSQL custom databases (non-postgres), include ${{ defaults.app_name }}-pg-init Job and implement startup-safe/idempotent creation logic (readiness wait + existence check before create).
For application-specific database compatibility, include an initContainer or startup gate that idempotently creates or repairs required views, aliases, indexes, extensions, privileges, role search paths, and legacy compatibility objects before the business container starts.
When an official runtime profile declares a database final-state requirement, include an initContainer gate that waits for the database and verifies the required extension or object before the business container starts.
Managed app main container command/args must stay close to the image's official entrypoint. Keep only official startup commands, Compose-native args, or a short exec wrapper; move file preparation, permission repair, database bootstrap, and compatibility self-healing into initContainers, Jobs, or ConfigMap scripts.
Shell wrappers in the main business container must exec the final process so signal handling remains correct.
Database bootstrap SQL must be safe under shell execution: prefer shell-level guard queries plus simple SQL, use single-quoted heredocs for psql variables, and avoid unguarded inline DO $$ blocks.
psql -c must not contain :'var' psql variable syntax; use heredocs for SQL that needs -v interpolation.
Database-specific constraints
Database services must use KubeBlocks Cluster resources, not application Deployment or StatefulSet workloads. StatefulSet is allowed for stateful application components only, never for PostgreSQL/MySQL/MongoDB/Redis/Kafka database services.
Database client images may be used in app initContainers and init/migration/bootstrap Jobs for readiness and bootstrap gates.
PostgreSQL RBAC requires app.kubernetes.io/instance and app.kubernetes.io/managed-by labels.
Every KubeBlocks database Cluster must include kb.io/database, sealos-db-provider-cr, and clusterdefinition.kubeblocks.io/name labels; sealos-db-provider-cr must equal metadata.name so dbprovider can list and classify the database. Related Pods, Services, and OpsRequests should carry app.kubernetes.io/instance=<database name> for detail views.
PostgreSQL role wildcard permission requirement remains as defined in current spec.
MongoDB cluster must follow upgraded structure (componentDef: mongodb, serviceVersion: 8.0.4, labels kb.io/database and app.kubernetes.io/instance).
MySQL cluster must follow upgraded structure (kb.io/database: ac-mysql-8.0.30-1, clusterDefinitionRef: apecloud-mysql, clusterVersionRef: ac-mysql-8.0.30-1, tolerations: []).
Redis cluster must follow upgraded structure (componentDef: redis-7, componentDef: redis-sentinel-7, serviceVersion: 7.2.7, main data PVC 1Gi, topology replication).
Database cluster component resources must use limits(cpu=500m,memory=512Mi) and requests(cpu=50m,memory=51Mi) unless source docs explicitly require otherwise.
All managed workload container resources must use the Sealos resource ladder: limits.cpu only 100m/200m/500m/1/2/3/4/8, limits.memory only 128Mi/256Mi/512Mi/1024Mi/2048Mi/4096Mi/8192Mi/16384Mi, and requests must be derived from limits by dropping the last numeric digit (500m→50m, 512Mi→51Mi, 1→100m, 1024Mi→102Mi, 4096Mi→409Mi). Do not invent non-ladder values, and never use 2G/4G/8G/16G because Sealos Template API quota preview can parse bare G memory as 0.
Do not add, delete, or change existing ephemeral-storage resource fields during existing-template updates unless runtime evidence identifies ephemeral storage pressure; preserve the original requests/limits values while tuning CPU and memory.
Secret naming:
MongoDB: ${{ defaults.app_name }}-mongo-mongodb-account-root (or ${{ defaults.app_name }}-mongodb-mongodb-account-root when the MongoDB cluster name uses -mongodb)
Redis: ${{ defaults.app_name }}-redis-redis-account-default (legacy ${{ defaults.app_name }}-redis-account-default may be accepted for backward compatibility)
Do not use legacy naming outside supported exceptions.
Baseline runtime defaults
Unless source docs explicitly require otherwise, use this lightweight app ladder entry as the initial personal low-load candidate:
container limits: cpu=200m, memory=256Mi
container requests: cpu=20m, memory=25Mi
revisionHistoryLimit: 1
automountServiceAccountToken: false by default; set it to true only when the application has explicit Kubernetes API/service account token requirements, evidenced by Kubernetes integration settings, serviceAccountName, or a sealos.io/service-account-token-reason workload annotation.
If a workload emits PodSecurity admission warnings and the image runs as a non-root user, add the restricted-compatible security context before reporting the template ready.
Static generation cannot prove the final resource tier. Complete live resource validation before treating the candidate as the final template value.
Personal low-load resource validation
Apply the resource ladder independently to every application main container, sidecar, initContainer, and Job:
The final CPU and memory limits must be the lowest Sealos ladder tiers that pass role-specific personal low-load validation, while an explicit source hard minimum remains the lower bound.
Tune CPU and memory separately, one ladder step at a time, and use a fresh rollout or cold execution for every candidate.
A passing long-running workload must complete cold start, become Ready, complete registration or login when applicable, complete at least two representative low-load actions, and remain stable for 60 seconds with zero OOMKilled terminations, restarts, readiness flaps, or resource-related timeouts.
A passing one-shot initContainer or Job must complete successfully from a cold run and allow every dependent workload to become Ready.
If a lower tier fails any acceptance signal, use the next passing tier and repeat final validation from a fresh rollout.
Treat observed CPU and memory peaks and utilization percentages as diagnostic evidence; acceptance failures trigger tier promotion.
Keep requests derived from limits according to the Sealos resource ladder.
In-container browser / remote desktop validation
Apply browser-specific validation only to containers that run Chrome, Chromium, VNC, WebRTC desktop, Xvfb, Selkies, noVNC, Kasm, or a similar remote-desktop stack; browser-accessed web applications such as Langflow use the general personal low-load policy.
Exercise cold start through readiness, a lightweight page, a real or medium page, an interactive or search action, and the 60-second stability window.
For Chrome + Xvfb + Selkies with a 4K maximum display, start validation at limits(cpu=200m,memory=1024Mi) with derived requests(cpu=20m,memory=102Mi), then test adjacent ladder tiers under the same acceptance contract.
Defaults vs inputs
defaults for generated values (app_name, app_host, random passwords/keys).
inputs only for truly user-provided operational values (email/SMTP/external API keys, etc.).
Classify the selected release's account flow as functional first-user signup, mandatory bootstrap credentials, or optional root reconciliation before defining administrator inputs.
When functional first-user signup is available and optional deploy-time administrator credentials have startup-fatal constraints beyond the Template input schema, use signup and omit administrator/root inputs plus their bootstrap env/config injection.
When mandatory bootstrap credentials are deployer-selected, declare the documented username or email and password fields in spec.inputs as required inputs with no default, describe the exact upstream constraints in English, validate the collected values before Template API deployment, and use the same values for live login. Keep database credentials on KubeBlocks secrets.
When mandatory bootstrap credentials are generated by the supported runtime, construct and validate the exact documented format deterministically, omit administrator inputs unless user selection is documented, retain the resolved credential through a Secret or live runtime source, and use it for redacted live login.
Prefer functional first-user signup over optional root reconciliation. Preserve mandatory bootstrap flows such as Frappe when the selected source requires them.
Every ${{ inputs.<name> }} reference in a template artifact must have a matching spec.inputs.<name> declaration in the same Template CR.
Every spec.defaults.<name>.value and every present spec.inputs.<name>.default in a Template CR must deserialize as a YAML string, regardless of the declared input type; quote numeric-, boolean-, and null-like scalars, while omitting default remains valid for required inputs.
inputs.description must be in English.
Startup-critical generated defaults must satisfy the application's documented validation with deterministic format construction. Use generated mandatory bootstrap only when the selected runtime documents deterministic generation and the resolved credential remains available through a Secret or live runtime source for redacted login. Keep first-user signup credentials in the registration flow.
Classify startup configuration-validation exits and repair the account-flow contract before resource tuning. Preserve every user-provided credential byte-for-byte between pre-deploy validation and live login.
For application-level optional object storage documented by the official source, use a boolean input (for example enable_s3_storage) and test with inputs.<name> === 'true'. Resolve provider/backend/type/mode/driver selection during conversion and keep those selectors out of spec.inputs.
The false branch of an optional object-storage input must configure the storage-disabled/local mode documented by the official source.
Optional-managed database contracts marked with docker-to-sealos.database-mode: optional-managed must share one boolean condition across the managed Cluster plane and database wiring while defining the documented SQLite or local false branch.
(CI / one-shot) python scripts/quality_gate.py --require-topology-evidence --artifacts /abs/path/template/<app-name>/index.yaml,.sealos/topology-evidence/<app-name>.yaml or DOCKER_TO_SEALOS_ARTIFACTS=/abs/path/template/<app-name>/index.yaml,.sealos/topology-evidence/<app-name>.yaml python scripts/quality_gate.py --require-topology-evidence (without explicit artifacts, it scans template/*/index.yaml and .sealos/*evidence*.yaml; set DOCKER_TO_SEALOS_ALLOW_EMPTY_ARTIFACTS=1 only for dev/debug without artifacts)
Live deploy acceptance: after sealos-deploy creates the app, verify the actual App URL, login/setup flow for web apps, recent logs, a documented API negative route or unique missing static asset without noisy traceback logs, expected database objects, and full resource footprint before reporting success.
check_consistency.py is registry-driven. Keep references/rules-registry.yaml in sync with implemented rules.
Registry rule entries support severity and optional scope.include_paths metadata.
Output Contract
When conversion is complete, provide:
brief conversion summary
target file path (template/<app-name>/index.yaml)
complete template YAML
key decisions only where ambiguity existed
Do not create or output README content in this skill. README generation is delegated to another skill.
supports --kompose-mode auto|always|never (always is default) to reuse kompose convert workload shapes
emits template/<app-name>/index.yaml
scripts/test_compose_to_template.py
regression tests for compose conversion behavior
scripts/check_consistency.py
registry-driven consistency validator
scripts/test_check_consistency.py
regression tests for validator behavior
scripts/check_must_coverage.py
validate MUST bullet coverage mapping against registry rules
scripts/test_check_must_coverage.py
regression tests for MUST coverage validator
Edge Policies
Never ask users for missing fields; infer from compose/docs and platform conventions.
Keep App resource in spec.data.url format; never use spec.template.
Keep App resource spec.displayType: normal and spec.type: link; do not infer alternative enum values.
Keep business-env, object storage, and DB-secret policy consistent with MUST rules.
Prefer square/circular icon-first logo assets (app icon/favicon/avatar) and avoid rectangular wordmark/text logos.
Prefer Sealos-managed ingress over bundled edge proxies: if a Traefik gateway is only acting as ingress/front-proxy and at least one business service exists, do not emit Traefik workload resources.
Prefer gateway TLS termination in Sealos Ingress over in-container TLS: for dual-port HTTP/HTTPS workloads, keep HTTP service port and remove redundant HTTPS/certificate mounts unless official docs require HTTPS backend.
Prefer WebSocket Ingress for public ws://, wss://, CDP/Chrome DevTools, game socket, and WebSocket-named ports/services; use backend-protocol: WS with 3600 read/send timeouts.
Never create template/<app-name>/README.md or template/<app-name>/README_zh.md; only keep README URL references inside index.yaml when required by the template schema.
Prefer fixing references/examples over adding exceptions when conflicts appear.
Use official Kubernetes installation docs/manifests to refine runtime semantics while retaining the selected source topology.
If the project mentions Frappe, ERPNext, HRMS, or bench, load references/frappe-bench.md before generating app workloads.
