Terraform

Terraform — Infrastructure as Code Writer

Writes Terraform code following strict module abstraction, file structure, and quality standards. Every module must represent a meaningful level of abstraction, use the latest stable releases, and pass format/init/validate before any commit.

Input Format

/terraform <task description>

• Describe the infrastructure to create, modify, or refactor
• Include the target provider (AWS, GCP, Azure, etc.) if known
• Include any specific constraints or existing module paths if relevant

Core Principles

Module Abstraction

Modules must represent meaningful levels of abstraction — never thin wrappers around a single resource. A module's value is in encoding an opinionated, complete infrastructure pattern that a caller treats as a single concept.

Wrong: A module that only wraps aws_s3_bucket and passes through a handful of arguments.

Right: A module that creates a bucket with server-side encryption, versioning, lifecycle rules, and optional replication — a complete, reusable pattern with sensible defaults.

Every module must have low complexity at its own level. If main.tf is growing hard to scan, delegate that complexity to sub-modules or shared modules. Each level should be easy to comprehend on its own without tracing through many moving parts.

Module Hierarchy

Terraform modules form a hierarchy. Each level has a defined responsibility:

The rule: if reading main.tf at any level requires tracking many moving parts to understand what it does, that level has too much complexity. Extract a sub-module.

File Structure

Every module uses this structure:

module-name/
├── main.tf        # Resource definitions and module calls
├── variables.tf   # Input variable declarations with descriptions and types
├── outputs.tf     # Output value declarations with descriptions
└── versions.tf    # terraform block with required_version and required_providers

Add files beyond this set only when a distinct logical concern warrants separation:

• locals.tf — when local value expressions are extensive
• data.tf — when many data source lookups would clutter main.tf

File naming rules (strictly enforced):

• Never prefix files numerically (01-network.tf, 02-compute.tf — forbidden)
• Never name files after infrastructure types (ec2.tf, rds.tf, vpc.tf, iam.tf — forbidden)
• File names describe logical role in the module, not the resource types inside them

Step 1: Pin to the Latest Stable Versions

Always use the latest stable release of Terraform and every provider. Never pin to an old version unless the user explicitly requires it.

Before writing any .tf file, look up the current versions:

1. Check the latest Terraform release at https://github.com/hashicorp/terraform/releases — use the highest non-prerelease version tag.
2. Check each required provider in the Terraform Registry at https://registry.terraform.io/browse/providers — confirm the latest published version.
3. Write versions.tf with those versions before writing any other file.
4. Note whether the latest versions match what you were trained on — this determines whether you need to consult provider documentation in Step 2.

versions.tf template:

terraform {
  required_version = "~> <latest_major.minor>"

  required_providers {
    <provider> = {
      source  = "<namespace>/<provider>"
      version = "~> <latest_major>"
    }
  }
}

Use ~> (pessimistic constraint operator) to allow patch updates within the pinned minor version for required_version, and patch/minor updates within the pinned major version for providers.

Step 2: Verify Resource and Data Source Arguments Are Current

Resource and data source arguments must match the current provider schema for the version pinned in versions.tf. Arguments get added, renamed, or deprecated between provider versions — using stale attribute names causes validation failures or silent misconfiguration.

1. If the latest provider version matches what you were trained on: you can write resources directly using your existing knowledge of the schema. No additional documentation lookup is needed.
2. If the latest provider version is newer than your training data: consult the provider documentation for every resource and data source you write before using any attributes:
   • Terraform Registry: https://registry.terraform.io/providers/<namespace>/<provider>/latest/docs/resources/<resource_type>
   • Fallback: the provider's GitHub repository under website/docs/r/
3. Use argument names exactly as they appear in the schema — do not guess or abbreviate.

Step 3: Write the Module

Apply these rules as you write each file:

main.tf

• Contains resource blocks and calls to sub-modules only
• No variable declarations, no output declarations, no provider configuration
• Should be readable top-to-bottom without needing to cross-reference other files constantly
• If it exceeds ~80 lines and is difficult to scan, delegate blocks to a sub-module

variables.tf

• Every variable must have a description
• Every variable must have an explicit type
• Provide default values only where a sensible, safe default exists
• Use validation blocks for values with known constraints (e.g., allowed regions, name length limits)

variable "environment" {
  description = "Deployment environment name (e.g. production, staging)."
  type        = string

  validation {
    condition     = contains(["production", "staging", "development"], var.environment)
    error_message = "environment must be one of: production, staging, development."
  }
}

outputs.tf

• Every output must have a description
• Export values that callers will need to wire modules together or reference externally
• Do not export internal implementation details that no caller would use

Dependency Injection

Modules must not resolve their own dependencies via internal data source lookups. The caller is responsible for providing all external resource references as input variables.

Wrong: A module that uses data "aws_vpc" "main" to look up the VPC by a name tag or environment convention.

Right: The caller passes var.vpc_id directly. The module declares a vpc_id variable and uses it without performing any lookup.

Apply this rule to any externally-owned resource: VPCs, IAM roles, KMS keys, Route 53 zones, etc. data sources are appropriate only for fetching provider-owned metadata (e.g. data "aws_availability_zones", data "aws_ami") or resources created within the same module call chain.

Polymorphism

A module should adapt its behaviour based on input variables rather than requiring callers to use different modules for different environments. Use conditional expressions and feature flags to produce distinct configurations from a single module.

variable "high_availability" {
  description = "When true, deploys a multi-AZ, multi-instance configuration."
  type        = bool
  default     = false
}

resource "aws_instance" "this" {
  count = var.high_availability ? 3 : 1
  # ...
}

This keeps the module surface small and ensures that production and development environments exercise the same code paths.

Sub-modules

Create a sub-module when:

• A logical group of resources within a type module makes main.tf hard to scan
• The same pattern is needed in more than one place

Place sub-modules in a modules/ directory within the parent module:

type-module/
├── main.tf
├── variables.tf
├── outputs.tf
├── versions.tf
└── modules/
    └── sub-concern/
        ├── main.tf
        ├── variables.tf
        └── outputs.tf

Step 4: Pre-Commit Workflow

Run these three commands in order inside every module directory that was created or modified. Do not commit until all three pass cleanly.

4a — Format

terraform fmt -recursive

Apply all reported formatting changes. Re-run until the command exits with no output (no changes needed).

4b — Init (no backend)

terraform init -backend=false

Downloads providers and validates the required_providers block without configuring a remote backend. Required before validate can run.

If init fails due to a version constraint conflict, re-check the registry for current versions and update versions.tf.

4c — Validate

terraform validate

All modules must pass with Success! The configuration is valid. — no errors, no warnings. Fix every issue before committing.

4d — Commit

Only after all three commands pass cleanly across every modified module directory.

Module Abstraction Checklist

Work through this before considering any module done:

• Does the module represent a meaningful abstraction, not a thin resource wrapper?
• Can you read main.tf top-to-bottom and understand what it does without deep investigation?
• Are all files named without numeric prefixes or infrastructure-type names?
• Is versions.tf present with the latest stable versions looked up from the registry?
• Did you fetch and verify provider docs for every resource and data source written?
• Are all variables typed and described?
• Are all outputs described?
• Does complex logic inside this module delegate to a sub-module rather than bloating main.tf?
• Does the module receive all external resource references as input variables rather than resolving them via internal data source lookups?
• Does the module use conditional logic or feature flags to adapt behaviour, rather than requiring separate modules per environment?
• Does the module avoid relying on naming conventions, environment names, or global state to locate resources?
• Have fmt, init -backend=false, and validate all passed cleanly?

Error Handling