1. 01Infrastructure

   Write it next to the code, in the same language

   Resources are typed: over four thousand of them generated from the providers’ own schemas, so the editor knows every field. Twenty-one cloud-agnostic abstractions resolve to the provider you name, which makes the cloud a variable rather than a rewrite.

2. 02Infrastructure

   Plan in the branch, review with the code

   The plan runs on the branch server with the environment you name. It lands in the change request beside the code diff, so the reviewer sees the node pool grow in the same place they see why.

3. 03Infrastructure

   The agent’s loop: plan freely, apply by policy

   An agent can verify and plan in its own branch and put the plan in its change request like any other diff. Whether it may apply is the run’s permission policy: a headless run that is not allowed to is denied rather than left hanging, an interactive one asks you.

4. 04Infrastructure

   Apply from the same branch, and the app learns the outputs

   State is yours: local or HTTP, encrypted and locked. Outputs flow back into the build as typed environment variables, so the service that needs the queue URL gets it from the plan that made the queue.

Coming soon: Kubernetes resources deploy through the k8s_deploy rule today, not the infra planner

The full reference lives in the docs.