Aspire Deployment

Aspire Deployment

Use this skill when the task is to publish, preview, validate, deploy, or tear down an Aspire application deployment. This skill owns Aspire deployment routing. Do not start with a generic Azure, Docker, Kubernetes, Helm, or Bicep workflow until you have checked whether the workspace is an Aspire app.

Aspire deployment starts from the AppHost model. Treat aspire deploy, aspire publish, aspire destroy, aspire do, and the deployment environment resources in the AppHost as the primary path. Use stable SDK/hosting packages from the project's release family and matching preview builds for preview-only integrations such as Kubernetes and Radius.

Keep this as one skill with target-specific references. Load only the reference files that match the target you discover or the user requests.

Routing precedence

This skill wins over generic cloud deployment skills when both conditions are true:

1. The user asks to deploy, publish, generate deployment artifacts, create Bicep/Helm/Compose/CDK output, host on Azure or AWS, deploy to Azure, AWS, Kubernetes, Docker Compose, or Radius, tear down deployed resources, or validate a deployment.
2. The workspace has Aspire markers:
   • Aspire workspace configuration
   • C# or TypeScript AppHost files
   • an AppHost project
   • AppHost code using Aspire distributed application builder APIs

If Aspire markers are present but this skill was not automatically invoked, switch to this skill before continuing. Prefer Aspire CLI commands such as aspire ls, aspire config list, aspire ps, and aspire describe for workspace orientation.

Guiding principles

Use Aspire docs before changing deployment code

Before adding target packages, editing the AppHost, or using an unfamiliar deployment API, use Aspire docs:

aspire docs search "deploy with Aspire"
aspire docs search "Docker Compose deployment"
aspire docs search "Kubernetes deployment"
aspire docs search "Azure Container Apps deployment"
aspire docs search "Azure App Service deployment"
aspire docs search "Azure Kubernetes Service deployment"
aspire docs search "Radius deployment"
aspire docs get "deploy-to-azure-kubernetes-service-aks"
aspire docs get "<slug-from-search-results>"

When you need exact C# or TypeScript API shape, use API docs too. Search both languages when you are not sure which AppHost language the repo uses:

aspire docs api search "<deployment API or concept>" --language csharp
aspire docs api search "<deployment API or concept>" --language typescript
aspire docs api get "<id-from-api-search>"

Do not invent package names, builder methods, overloads, or deployment commands. API shapes differ between C# and TypeScript AppHosts.

Prefer Aspire-native deployment

Deploy the resources declared in the AppHost, not the AppHost process itself. For Docker Compose, configure the Compose environment and let Aspire generate services for those resources; do not replace this with an AppHost Dockerfile or Compose service.

Use Aspire deployment targets and CLI commands first:

aspire publish --list-steps
aspire deploy --list-steps
aspire publish
aspire deploy
aspire destroy
aspire do <step>

Use target-specific tooling only after Aspire has generated artifacts or when the target docs call for it:

• Docker Compose: inspect generated aspire-output/docker-compose.yaml and .env*; Aspire can also run docker compose up through aspire deploy.
• Kubernetes: inspect generated Helm chart output; use Helm/kubectl when applying published artifacts yourself.
• Azure: use aspire add <azure-target>, aspire publish, and aspire deploy through the AppHost deployment environment.
• AWS: use aspire add aws to add the integration, inspect generated CDK/CloudFormation output, and follow the AWS integrations repository guidance.
• Radius: treat Aspire.Hosting.Radius as preview. Use integration/docs search to verify the package and API, then model AddRadiusEnvironment(...).WithNamespace(...).

Ask where to deploy only when ambiguous

Do not ask for target selection when the user already chose a target such as Docker Compose, Kubernetes, Azure Container Apps, Azure App Service, Azure Kubernetes Service (AKS), or AWS. Use the chosen target and continue with its reference.

If the user did not explicitly choose a deployment target and the AppHost does not already contain exactly one deployment environment, ask where they want to deploy before adding integrations, editing the AppHost, publishing artifacts, or deploying. Use a single multiple-choice question:

Where do you want to deploy this Aspire app?

Show these choices:

Do not offer preview Radius as a default choice. Use it when the user explicitly requests Radius or the AppHost already contains a Radius environment.

If the user says only "Azure", ask again with just the Azure choices: Azure Container Apps, Azure App Service, or Azure Kubernetes Service (AKS). If the AppHost already contains exactly one deployment environment and the user did not ask to change targets, use that target and tell the user what was detected.

Ask before creating cloud resources when intent is not explicit

Cloud deploys can create billable resources. If the user asked for a plan, preview, validation, or "make this deployable", stop after the deployment plan/artifacts and ask before running the command that provisions resources.

If the user explicitly asked to deploy now, continue through preflight and deployment, but still surface any target choice, subscription/resource group ambiguity, or missing parameter decisions before provisioning.

Keep Azure deployment Aspire-native

The Azure deployment path in this skill is aspire add <azure-target>, AppHost environment configuration, aspire publish, and aspire deploy. Do not route Azure deployment work through a separate Azure deployment tool or generated infrastructure workflow.

Keep Aspire packages on one release family

Before publish or deploy, verify that Aspire.AppHost.Sdk and the hosting packages use compatible versions in the project's selected release family. Preview-only integrations need compatible preview builds selected by aspire add or integration docs. The CLI version does not make incompatible hosting packages safe; mixed graphs can fail at startup with MissingMethodException or TypeLoadException.

Migrate obsolete 13.5 surfaces while editing deployment code:

• use .Services instead of hosting context .ServiceProvider;
• use AddConnectionString instead of PublishAsConnectionString; and
• migrate deprecated Aspire.Hosting.GitHub.Models resources to Azure AI Foundry.

Default workflow

1. Orient to the Aspire workspace.
   • Start with aspire ls to list AppHosts in the current scope, then use aspire.config.json, AppHost project files, or aspire ps if more context is needed.
   • If no AppHost exists, stop deployment work and invoke the aspireify skill to initialize/wire the AppHost before continuing.
   • Identify C# vs TypeScript AppHost.
   • Prefer Aspire CLI commands for discovery and state inspection.
2. Clarify or infer the deployment target.
   • If the user named Docker Compose, Kubernetes, Azure Container Apps, Azure App Service, Azure Kubernetes Service (AKS), or AWS, load that target reference without asking again.
   • If they only said "deploy", inspect existing AppHost target environment resources.
   • If exactly one target environment already exists, use it and state what was detected.
   • If multiple targets exist, none exists, or the user says only "Azure", ask where to deploy using the choices above.
3. Load target and app-type references.
   • Docker Compose: references/docker-compose.md
   • Kubernetes and Azure Kubernetes Service (AKS): references/kubernetes.md
   • Azure Container Apps/App Service/Azure Kubernetes Service (AKS): references/azure.md
   • AWS: references/aws.md
   • Radius preview: use current Aspire docs and API search for Aspire.Hosting.Radius; do not guess experimental API shapes.
   • JavaScript app resources: references/javascript.md
   • CI/CD or GitHub Actions automation: references/cicd.md
4. Use Aspire docs search for current guidance.
   • Search and get the target deployment docs.
   • Search API docs before editing AppHost code.
5. Apply the target code changes.
   • Run the target's aspire add ... command if the integration is missing.
   • Add the deployment environment resource to the AppHost.
   • Do not add explicit compute-environment assignment for the common single-environment case. Only disambiguate when the AppHost has multiple deployment environments. In C# this is usually WithComputeEnvironment(...); for TypeScript AppHosts, verify the current language-specific docs before assuming an equivalent.
   • Add only the target-specific customization APIs the deployment needs, such as endpoint exposure, Helm settings, Compose file customization, Azure site/container app customization, or AWS publish target overrides.
6. Preflight the deployment model.
   • Confirm the target integration package exists in the AppHost.
   • Confirm the AppHost has the target environment resource.
   • Confirm compute resources are assigned to the target environment only when multiple compute environments exist. A single compute environment is the common case and can be inferred.
   • Inventory parameters, secrets, connection strings, external endpoints, container registries, and target-specific prerequisites.
   • For Azure or AWS, confirm auth, target account/subscription, region/location, and resource group/stack context.
7. Preview before applying.
   • Run aspire publish --list-steps or aspire deploy --list-steps.
   • Use aspire publish -o <scratch-or-output-path> when artifact review is requested.
   • Treat published artifacts as a preview/handoff. aspire deploy resolves values and applies the deployment from the AppHost model; it does not consume a previously published output directory.
   • Summarize resources, endpoints, parameters, secrets, identities, and generated artifacts.
8. Deploy or hand off.
   • Run aspire deploy when the user asked to deploy and preflight is complete.
   • Run a named step with aspire do <step> only when the user asked for a specific pipeline step.
   • For published artifacts, explain the target-native apply step.
9. Destroy only when explicitly requested.
   • Run aspire destroy to execute the selected AppHost/environment's target destroy pipeline.
   • Confirm the AppHost, environment, target account/subscription/cluster, and destructive intent before running it.
   • Use --yes only when the user or CI workflow already made teardown intent explicit.
   • Prefer aspire destroy over target-native delete commands unless you are troubleshooting failed teardown or cleaning up unmanaged leftovers.
10. Verify the outcome.

• Use target output, aspire describe, cloud CLI, Docker Compose, kubectl, or endpoint checks appropriate to the target.
• After destroy, verify target resources are removed or record any leftovers that require manual cleanup.