Cross-tool actions

Getting keys is layer 1. The payoff is layer 2 and 3: with a root grant in hand, the agent maps the actions now possible for your repo and does the safe ones without you.

What "actions" are

The agent calls mapActions and the cockpit renders a living panel of repo-specific and cross-tool steps, for example:

• a Neon branch per environment,
• wire Infisical → Cloudflare Pages bindings,
• set a Workers binding,
• point a domain at Cloudflare,
• create the R2 bucket your code already references.

Each Action carries a why, its prerequisites, a danger level, and the wizard that runs it. It's schema-validated at the boundary — malformed is rejected.

Directing the panel

The action list is living, not fixed. Talk to the agent in the dashboard chat:

• "also set up Stripe" → it adds the action,
• "skip the R2 bucket" → it removes it,
• "add a staging env" → it adjusts and re-renders.

You steer; it re-maps. Ask "why does this need that scope?" and it answers in chat, grounded in your repo — without leaving the cockpit.

Example: domain → Cloudflare

The domain→CF executor points a domain at Cloudflare. A nameserver swap is destructive, so it is hard-confirmed — never one-click. The flow is: the agent maps the action, you approve, you clear the confirm gate, then the daemon executes with the stored creds and streams status back.