Getting started

1. Get approved. Message the bot once — that request lands in the operator’s queue. Approval creates your personal machine, and everything you set up meanwhile is waiting for it: /sparkyard github and /sparkyard secret already work while you’re in the queue.

2. Connect an agent. Run /sparkyard connect — it lists the four and shows which you already have. Sign in to whichever you want:

   agent	command	what happens
   Claude	/sparkyard connect claude	A sign-in link. Approve it anywhere, paste the short code back into the form.
   ChatGPT (Codex)	/sparkyard connect chatgpt	Paste an API key from platform.openai.com into a private form.
   Gemini	/sparkyard connect gemini	Paste a key from aistudio.google.com.
   GitHub Copilot	—	Nothing to do: it signs in as you on GitHub, which is step 3.

   Whatever you connect is sealed for you alone, and your turns bill your account and nobody else’s. Claude’s token is minted inside your own VM and verified against Anthropic before it is stored; keys are never pasted into a channel, only into a form.

   You can skip Claude if your admin has set an org-wide Anthropic key with /sparkyard org-key — your Claude turns bill that until you connect an account of your own, which always wins. There is no equivalent for the other three: those need a key of your own.

   Connected more than one? //agent in a thread picks who answers it; /sparkyard agent sets your standing default.

3. Connect GitHub — /sparkyard github. A device-flow code to type at github.com/login/device. Your pushes and pull requests are then authored as you, with a short-lived token scoped to the repositories the Sparkyard GitHub App is installed on. First time? The message includes an Install the app on repos button — that consent screen is the one step only you can click.

4. Say hello. Mention the bot in a channel or DM it. Name a repository with repo:name, or name none and tell the session what to clone once it asks — deciding the repo is part of the conversation.