Terraform Patterns

Write predictable, secure Terraform

Predictable infrastructure. Secure state. Modules that compose. No drift.

Module structure (standard layout)

modules/<name>/
  main.tf        # resources only — no provider config
  variables.tf   # typed inputs with descriptions and validation
  outputs.tf     # only what callers need
  versions.tf    # required_providers + minimum version constraints

Environments (envs/prod/, envs/staging/) consume modules via module blocks — never paste resources directly into env files.

Non-negotiable rules

Steps

1. Read the lore + existing modules. search_lore for state backend, naming conventions, and existing module registry. Extend existing patterns; don't introduce a competing module style.
2. Design the module interface. Variables: typed + validated + described. Outputs: only what callers need — not every internal resource attribute. One module = one coherent infrastructure concern.
3. Implement resources. Follow least-privilege on IAM. Encrypt at rest and in transit by default. Tag all resources with environment, team, and managed-by = terraform.
4. Configure the backend. Remote state with locking. Workspace or path-based isolation between environments.
5. CI/CD integration. terraform fmt + terraform validate on every PR; terraform plan as a required check; apply only on merge to main via a protected pipeline.
6. Security review. Check for: hardcoded credentials, 0.0.0.0/0 ingress, public S3 buckets, overly permissive IAM. Use tfsec or checkov in CI.
7. Verify. terraform plan shows expected change set only; terraform apply completes without errors; spot-check resources in cloud console; record evidence.

Build / Test

• terraform fmt -check -recursive — format violations fail CI.
• terraform validate — catches provider config errors before plan.
• tfsec or checkov — static security analysis; block on HIGH/CRITICAL findings.
• terraform plan output reviewed and linked in PR before apply.

Review checklist

• Module interface clean — typed variables with descriptions; outputs are the minimum callers need.
• Remote state + locking — no local state; lock configured.
• No hardcoded credentials — secrets from environment or vault only.
• IAM least privilege — no * actions or resources without documented justification.
• Resource tagging — environment, team, managed-by on every resource.
• Security scan clean — no HIGH/CRITICAL findings from tfsec/checkov.

Capture lore

State backend configuration, naming conventions, tagging policy, and approved module registry are high-value Terraform lore — call suggest_lore with tags: [terraform, infra, iac].