Component Development

SkillDev tools

Atmos core component development: adding or changing native component types, component registry providers, commands, stack schema, docs, examples, DAG/affected behavior, auth, hooks, source/provisioning, and tests

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Component Development skill

What this skill tells your AI

The instructions your AI receives, as published by cloudposse/atmos in .claude/skills/component-development/SKILL.md and read by ahel’s review.

Use this skill when adding or changing an Atmos native component type or shared component behavior.

Start With The Existing Contract

Inspect comparable component types before designing:

rg -n "ComponentType|SectionName|components\\..*base_path" pkg/config pkg/schema
rg -n "component.Register|ComponentProvider|Execute\\(" pkg/component cmd internal
rg -n "process.*ComponentsIndexed|ExecuteGraph|dependencies.components" internal pkg

For stack-surface decisions, compare Terraform, Helmfile, Packer, and Ansible. Reuse shared sections when they already exist:

  • metadata
  • vars
  • env
  • settings
  • hooks
  • auth
  • dependencies
  • generate
  • source
  • provision

Do not invent nested settings for component-owned inputs. If an instance is components.<type>.<name>, component-specific inputs belong directly under that instance, e.g. components.kubernetes.argocd.paths, not settings.kubernetes.paths.

Required Wiring

For a new component type, update all applicable surfaces in one PR:

  • Constants: pkg/config/const.go
  • CLI/default config structs: pkg/schema/schema.go
  • Defaults: pkg/config/default.go
  • Config path resolution: pkg/config/config.go, pkg/utils/component_path_utils.go
  • Component registry provider: pkg/component/<type>/
  • Command package: cmd/<type>/
  • Root side-effect imports: cmd/root.go
  • Source provisioning base path support: pkg/provisioner/source/
  • Hooks/events if lifecycle hooks are supported: pkg/hooks/
  • Affected detection: internal/exec/describe_affected_*
  • DAG/bulk execution if --all or --affected is supported
  • Embedded CLI usage examples: cmd/markdown/atmos_<command>*_usage.md
  • Runtime stack schemas: pkg/datafetcher/schema/atmos/manifest/1.0.json and pkg/datafetcher/schema/stacks/stack-config/1.0.json
  • LSP hints if stack authoring changes: pkg/lsp/server/
  • Docs: command docs, stack component docs, atmos.yaml config docs
  • Examples under examples/ for user-visible component types

If a public website schema file exists in the checkout, update it too. Do not assume it exists; verify with find website/static -path '*schema*'.

Command Behavior

Commands should be Atmos-native, not thin binary wrappers, when the implementation owns behavior.

  • Use the component registry provider rather than hard-coded command logic.
  • Use flags.NewStandardParser() for command-specific flags.
  • Preserve --identity and profile handling.
  • For bulk operations, allow zero positional component args with --all or --affected.
  • Reject ambiguous forms such as component arg plus --all.
  • Use dotted hook events such as before.<type>.<operation> and after.<type>.<operation>.
  • Add embedded usage examples for every public subcommand.

If a command name like kubectl or kustomize is used as a provider, be explicit whether it means CLI execution or compatible behavior. For SDK-backed providers, do not add a command override unless a binary is actually executed.

Auth

Thread Auth through stack processing and execution:

  • Parse global --identity into ConfigAndStacksInfo.Identity.
  • Use the same auth setup path as existing component commands when stack YAML functions or integrations need credentials.
  • Keep component-level auth: in the stack schema.
  • For Kubernetes-like SDK clients, create the client after Auth has prepared environment or kubeconfig integration output.
  • Prefer plain ambient for local examples that should respect the surrounding environment without prompting for provider login. Use concrete provider identities for examples that actually require a cloud or service-specific authentication flow.

DAG And Affected

If the component supports --all or --affected:

  • Build nodes from components.<type>.
  • Skip abstract, disabled, and locked components where applicable.
  • Use dependencies.components as the primary dependency source.
  • Keep legacy settings.depends_on only as compatibility.
  • Process in topological order.
  • For --affected, run affected detection first, filter to the component type, then graph-filter selected nodes.
  • Include dependencies by default so prerequisites run before selected nodes.
  • Make --include-dependents explicit for downstream components.

Affected detection must include both implementation path changes and stack config changes. For new component-specific top-level sections, add affected reasons for those sections.

Schemas And Docs

Schema and docs are part of the feature, not follow-up polish.

Update stack schemas with the exact component shape. Validate Atmos-owned fields strictly, but keep domain resources permissive when a complete third-party schema would be brittle. Example: validate Kubernetes manifest entries have apiVersion and kind, but do not model the entire Kubernetes API.

Docs must cover:

  • atmos.yaml configuration
  • Stack component configuration
  • Every command and flag
  • Reusable sections supported by the component
  • Auth behavior
  • Hooks and lifecycle events
  • Dependencies, --all, and --affected
  • Examples for file paths, inline manifests/config, render/generate behavior, source/provisioning when supported

Use the local docs skill for website docs conventions.

Command and component docs must explain the user outcome before implementation details. Do not open intros with mechanical phrasing such as "Use this command to..." or with internal architecture such as providers, SDKs, DAGs, or stack processing unless that detail changes how the user should operate the feature. Lead with what the user can accomplish, when they should reach for the command or configuration page, and the next action they can take.

Examples

Examples should be runnable and reflect the intended user model:

  • Use local infrastructure like k3s when possible.
  • Keep command examples working without cloud credentials unless the example is explicitly cloud-auth focused.
  • If Auth is part of the feature, include a local identity path and a commented real-cloud integration example when useful.
  • Add separate examples for materially different providers or modes.

Validation

Run focused tests first:

go test ./pkg/component/<type> ./cmd/<type>
go test ./pkg/config ./pkg/schema ./pkg/hooks ./pkg/provisioner/source
go test ./internal/exec ./pkg/lsp/server
jq empty pkg/datafetcher/schema/atmos/manifest/1.0.json
jq empty pkg/datafetcher/schema/stacks/stack-config/1.0.json

For docs:

pnpm --dir website exec prettier --check "docs/**/*.{mdx,json}"

For examples:

(cd examples/<name> && atmos validate stacks)
(cd examples/<name> && atmos <type> render <component> -s <stack>)

If a validation failure is pre-existing or requires external services, say that explicitly and include the narrow command that failed.

Signals

GitHub stars
1k
Forks
175
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
component-development
Source
github.com/cloudposse/atmos