AgentSkillsCN

deploy

智能调度，将应用部署至多个集群。

SKILL.md
--- frontmatter
name: deploy
description: Deploy an app to multiple clusters with smart placement
allowed-tools: mcp__plugin_kubestellar-deploy_kubestellar-deploy__*

Your task

Deploy a workload to clusters. You can specify target clusters explicitly or let kubestellar-deploy find matching clusters based on requirements.

  1. •Get the manifest from the user or help them create one
  2. •Ask about target clusters or placement requirements (GPU, memory, CPU)
  3. •Use deploy_app to deploy the manifest

Parameters

  • •manifest: The Kubernetes YAML manifest
  • •clusters: Optional list of target clusters
  • •gpu_type: Optional GPU type requirement (e.g., "nvidia.com/gpu")
  • •min_gpu: Optional minimum GPU count
  • •dry_run: Set to true to preview without applying

Smart Placement

If the user doesn't specify clusters, kubestellar-deploy can:

  • •Deploy to all clusters
  • •Filter to clusters with specific GPU types
  • •Filter to clusters with minimum CPU/memory
  • •Filter to clusters with specific node labels

Use find_clusters_for_workload and list_cluster_capabilities to help with placement decisions.

Examples

  • •"Deploy this nginx deployment to all clusters"
  • •"Deploy my ML model to clusters with GPUs"
  • •"Deploy this app to clusters with at least 16Gi memory"
  • •"Do a dry run of deploying this manifest"

Do not use any other tools besides the kubestellar-deploy MCP tools.