Terraform Best Practices

This skill provides guidance for writing reliable, secure Terraform configurations in RedC deployment scenarios. The recommendations here come from real-world experience with multi-cloud red team infrastructure — where a misconfigured state file or leaked credential can compromise an entire operation.

State Management

Use remote state because local state files are a single point of failure. If the file gets deleted, corrupted, or conflicts with another operator's changes, you lose track of what's deployed — which in red team scenarios means orphaned infrastructure you're still paying for.

• Store state in cloud-native backends (S3+DynamoDB for AWS, OSS for Alibaba, COS for Tencent)
• Enable state locking to prevent two operators from applying simultaneously
• Keep .tfstate out of version control — it often contains sensitive outputs like IP addresses and credentials
• Separate state per environment (dev/staging/prod) so a bad apply in dev doesn't corrupt prod state

Example — remote backend configuration:

terraform {
  backend "s3" {
    bucket         = "myteam-tfstate"
    key            = "prod/infra.tfstate"
    region         = "us-east-1"
    dynamodb_table = "tf-lock"
    encrypt        = true
  }
}

Module Design

Modules keep configurations DRY and composable. A well-designed module encapsulates one concern (e.g., "a VPC with public/private subnets") so you can reuse it across deployments without copy-pasting.

• Keep each module focused on a single responsibility — a "vpc" module shouldn't also create EC2 instances
• Use validation blocks on variables to catch bad input before plan time
• Output the values downstream resources need (IDs, IPs, endpoints, DNS names)
• Pin module versions in production (source = "git::...?ref=v1.2.0") to avoid surprise changes

Example — variable validation:

variable "instance_type" {
  type        = string
  description = "EC2 instance type for the deployment"
  default     = "t3.small"
  validation {
    condition     = can(regex("^(t3|t4g|m5|c5)\\.", var.instance_type))
    error_message = "Use t3, t4g, m5, or c5 family instances for red team workloads."
  }
}

Security

Red team infrastructure is high-value target material. A leaked credential or overly permissive security group doesn't just cost money — it can expose your operation.

• Never hardcode credentials in .tf files. Use environment variables (TF_VAR_*), a secrets manager, or RedC's profile-based credential injection
• Restrict security group ingress to the CIDRs you actually need. Opening 0.0.0.0/0 on port 22 invites scanners within minutes
• Enable encryption at rest for all storage resources (EBS, S3, disks) — it's usually free and prevents data exposure on decommission
• Prefer IAM roles attached to instances over embedding access keys, because roles auto-rotate and don't appear in state files
• Tag every resource with owner, purpose, and expires so forgotten infrastructure can be identified and cleaned up

Variables and Locals

Good variable hygiene makes templates reusable and self-documenting.

• Set sensible defaults for optional variables so users can deploy with minimal config
• Write a description for every variable — RedC's template system surfaces these to users
• Mark sensitive variables with sensitive = true to prevent their values from appearing in logs and plan output
• Use locals for computed values to avoid repeating expressions across resources

Example — locals for computed naming:

locals {
  name_prefix = "${var.project}-${var.environment}"
  common_tags = {
    Project     = var.project
    Environment = var.environment
    ManagedBy   = "RedC"
  }
}

Common Pitfalls

These are the mistakes that come up most often in RedC deployments:

Symptom	Likely Cause	Fix
"Could not load plugin" after adding a provider	Forgot terraform init	Run terraform init -upgrade
Resources created in wrong order	Missing implicit dependency	Add explicit depends_on
Plan shows unexpected changes	State drift from manual console changes	Run terraform refresh then review
Sensitive value visible in output	Forgot sensitive = true	Add it to the variable and output blocks
-target left partial state	Targeted apply skipped dependent resources	Apply the full plan without -target to reconcile