📢 CloudStack CMK Skill for AI Agents — Automating CloudStack with LLMs

[chunkyen]

I've been experimenting with using AI agents (Hermes / OpenCLAW) to manage CloudStack infrastructure, and I built a reusable OpenClaw skill that gives AI agents full access to CloudStack via cmk (CloudMonkey):

🔗 https://github.com/chunkyen/cloudstack-cmk-skill

There are other approaches out there — like the CloudStack MCP Server (https://github.com/phantosmax/cloudstack-mcp-server) which wraps individual APIs as predefined tools. But I went a different route: instead of hand-picking 45 APIs, this skill gives the agent access to the entire cmk CLI — all 870+ CloudStack APIs through a single interface.

The key advantage: Once you run cmk sync, the entire API surface is cached locally as JSON. The LLM can use jq to explore it and construct valid commands for any API — even newly added or obscure ones — without any code changes or manual tool definitions. No waiting for someone to write a new tool when CloudStack adds a feature.

Other highlights:

• Built-in approval workflows for destructive operations — though this is not a guarantee, it depends on the AI agent following instructions correctly
• Works with any cmk auth method (password, API key, SAML SSO)
• Supports sub-agents and background tasks for complex workflows

⚠️ Important disclaimer: This skill operates with whatever permissions are assigned to the cmk login. The agent cannot enforce restrictions beyond what CloudStack's own RBAC allows for that credential. If logged in as root admin, the agent can perform any operation including destructive ones. You are responsible for ensuring the cmk profile uses a credential with the minimum necessary privileges for your use case. The built-in confirmation prompts are a safety net, not a substitute for proper credential scoping.

This skill was a huge help in my broader project — it powered the testing and documentation of all four CloudStack Kubernetes Flavors (https://github.com/chunkyen/cloudstack-kubernetes-flavors). Being able to deploy VMs, manage networks, and troubleshoot clusters through an AI agent made the whole process significantly faster.

This is a project built out of passion for CloudStack. If you share that passion, give it a try. Issues and contributions welcome!