Terraform Changes

Terraform Changes

Use this skill whenever a change touches Terraform-managed infrastructure (clusters, databases, config). The goal is that every change is grounded in current state and has a known rollback path before it runs.

Read state first

1. Identify the resource and its environment. Never plan against the wrong environment.
2. Inspect current state — replica counts, instance class, whether the resource is stateful. A resource that holds data cannot be safely destroyed and recreated.
3. List downstream dependents. These become the plan's affected_services.

Read the plan diff

A Terraform plan reports each resource as one of:

• + create — additive, generally low risk.
• ~ update in-place — usually low to medium risk; safe to roll back by re-applying the prior value.
• -/+ replace — destroy then create. Treat as high risk on anything stateful; data is lost between destroy and create unless a snapshot exists.
• - destroy — removal. High risk on shared or stateful resources.

Map the diff to a ChangePlan.action: a replica change is scale, an in-place config change is apply, a -/+ or - on a real resource is destroy.

Safety rules

• Never destroy or -/+ replace a stateful resource without a verified snapshot in the rollback steps.
• Scale changes must stay within the resource's max_replicas. Scaling past headroom is not a low-risk change.
• Every plan must have ordered rollback_steps. If you cannot write a rollback, the risk is high and the change should be questioned, not just approved.
• Treat any production change as at least medium risk by default.

See references/risk-matrix.md for the action-to-risk mapping used to fill the ChangePlan.risk field.