---

name: stacks-and-integrations description: "Register, configure, and extend ZenML stacks, stack components, integrations, service connectors, materializers, orchestrators, step operators, artifact stores, container registries, image builders, and optional dependency boundaries." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

ZenML Stacks and Integrations

Use this sub-skill when a task mentions ZenML stacks, stack components, flavors, integrations, service connectors, materializers, orchestrators, step operators, artifact stores, container registries, image builders, Docker/Podman image builds, remote execution backends, cloud credentials, or missing optional SDK extras.

Route First

• Read references/component-patterns.md for stack registration, stack validation, component config/settings, service connector use, materializer contracts, and remote image-building requirements.
• Read references/integration-development.md before adding or changing an integration package, flavor, orchestrator, step operator, image builder, container registry, artifact store, or optional dependency boundary.
• Read references/troubleshooting.md when imports, extras, service connectors, stack validation, Docker, image builds, orchestrator IDs, or step-operator lifecycle behavior fail.
• Run scripts/check_optional_imports.py from an active ZenML checkout before finishing integration flavor changes that may import optional SDKs.

Boundaries

• Own stack infrastructure and integration authoring/import safety; for using DockerSettings, ResourceSettings, schedules, materializers, or stack settings inside user pipelines, cross-read ../pipeline-authoring/SKILL.md.
• Own component-level and integration-level validation rules; for repository-wide test selection, formatting, linting, documentation, dependency, or CI workflow choices, cross-read ../maintenance/SKILL.md.
• Use ../cli-and-client/SKILL.md for CLI/client resource listing, filters, project/server login, and stack-management commands when the task is usage-focused rather than implementation-focused.
• Do not duplicate server/store internals here; route FastAPI, SQLModel, migrations, RBAC, and store schema work to the server/store sub-skill.

Non-Negotiable ZenML Rules

• Keep optional third-party SDK imports out of integration flavor module top level; config classes in flavor files should use standard library, Pydantic, and ZenML core types only.
• Import implementation classes lazily inside implementation_class properties, and use TYPE_CHECKING blocks for optional implementation type hints.
• For remote orchestrators and step operators, validate remote artifact stores, container registries, required image builders, and remote server requirements before runtime submission.
• get_orchestrator_run_id() must return one stable ID for every step in a pipeline run and a unique ID across different runs; dynamic retry paths must preserve the same orchestration-environment ID.
• New step operators should implement submit(), get_status(), wait(), and cancel(); store backend job IDs immediately after submission so polling and cancellation can recover.