# Omega Cloud -- Using Your AI Assistant Skill (DRAFT)

With your handover pack you received `tenant-skill/`, a skill package for
Claude (Anthropic's AI assistant). Installed into your own project, it
teaches your assistant how to operate your Omega Cloud environment
correctly: which of your three accounts to use for which job, which
permission errors are deliberate platform boundaries, and how to behave
safely on a live environment. This page explains how to install it and
how to get the most out of it.

## What the skill is

A folder of plain-text instructions your AI assistant loads when you ask
it to do cloud work:

- `SKILL.md` -- the core: account selection, the boundary model,
  authentication, safety rules, and a routing table.
- `references/` -- detailed, on-demand pages the assistant loads per
  topic: day-2 operations (networks, VMs, volumes, load balancers),
  Kubernetes, CI/automation, and troubleshooting.

Your copy was instantiated for you at handover: it already contains your
short name, endpoints, and region. It contains NO credentials -- it tells
the assistant how to authenticate, never what your secrets are.

## Installing it

- **Claude Code** (terminal / IDE): copy the delivered `tenant-skill/`
  directory into your project as `.claude/skills/omega-cloud-tenant/`
  (the directory name must be exactly `omega-cloud-tenant`). The skill
  is picked up automatically; asking "what skills are available?" should
  list it.
- **claude.ai** (web): add the skill files to your project so they are
  part of its knowledge, or use your organization's skill-upload feature
  where available. Claude Code with terminal access is the better fit
  for hands-on cloud work, since the assistant can run the OpenStack
  CLI itself.

Give the assistant a working CLI environment: the OpenStack client
installed, your `clouds.yaml` configured per the CI/Automation
Integration Guide, and the delivered CA bundle on disk. Credential
secrets belong in your environment (or your secret store), never pasted
into the chat.

## Getting the best out of it

1. **Just describe the task.** "Create a VM reachable from our office
   network", "why did the pipeline get a permission error", "stand up a
   Kubernetes cluster for staging". The skill triggers on cloud work by
   itself; you do not need to name it.
2. **Let it read before it writes.** The skill instructs the assistant
   to audit current state with list/show commands before changing
   anything, and to ask your confirmation before anything destructive.
   Expect (and welcome) those confirmation questions.
3. **Expect it to respect boundaries.** If you ask for something the
   platform deliberately refuses -- raising your own quota, creating a
   cluster from an automation credential, granting `admin` -- a
   correctly working assistant will decline the workaround and point
   you at the request path instead. That is the skill doing its job.
4. **Never paste secrets into the conversation.** Passwords, credential
   secrets, kubeconfig contents, and private keys stay out of the chat.
   The assistant is instructed never to print them; help it by never
   supplying them in-line. When a password login is needed, it will
   have you enter it at a prompt.
5. **Keep the companions nearby.** The Handover Pack, Self-Service
   Guide, CI/Automation Integration Guide, and Acceptance Checklist are
   the authoritative long-form documents; the skill is the operating
   layer over them. If the assistant's answer and a delivered document
   ever disagree, the document wins -- and we would like to hear about
   it via your account contact.
6. **A good first test.** Ask the assistant to run the delivered
   read-only audit script (`bash scripts/tenancy-audit.sh` from your
   document pack). It proves authentication, the CA bundle, endpoint
   discovery, and the skill itself in one pass, and changes nothing.

## Keeping it current

The skill is versioned with your document pack. When we ship an updated
pack, replace the installed skill directory with the new copy -- do not
hand-edit it, or your edits will be lost at the next update. Requests to
change what the skill teaches go through your account contact like any
other document feedback.