terraform-skill

Terraform Skill for Claude

Diagnose-first guidance for Terraform and OpenTofu. Core file is a workflow; depth lives in references loaded on demand.

Response Contract

Every Terraform/OpenTofu response must include:

1. Assumptions & version floor — runtime (terraform or tofu), exact version, providers, state backend, execution path (local/CI/Cloud/Atlantis), environment criticality. State assumptions explicitly if the user did not provide them.
2. Risk category addressed — one or more of: identity churn, secret exposure, blast radius, CI drift, compliance gaps, state corruption, provider upgrade risk, testing blind spots.
3. Chosen remediation & tradeoffs — what was chosen, what was traded off, why.
4. Validation plan — exact commands (fmt -check, validate, plan -out, policy check) tailored to runtime and risk tier.
5. Rollback notes — for any destructive or state-mutating change: how to undo, what evidence to keep.

Never recommend direct production apply without a reviewed plan artifact and approval.

Never run terraform destroy (targeted or full) without first running terraform plan -destroy and showing the user every resource that will be deleted — including implicit dependents pulled in via locals or for_each. Get explicit confirmation before proceeding. Never use -auto-approve on destroy.

Workflow

1. Capture execution context — runtime+version, provider(s), backend, execution path, environment criticality.
2. Diagnose failure mode(s) using the routing table below. If intent spans categories, load both references.
3. Load only the matching reference file(s) — do not preload depth the task does not need.
4. Propose fix with risk controls — why this addresses the mode, what could still go wrong, guardrails (tests/approvals/rollback).
5. Generate artifacts — HCL, migration blocks (moved, import), CI changes, policy rules.
6. Validate before finalizing — run validation commands tailored to risk tier.
7. Emit the Response Contract at the end.

Diagnose Before You Generate

Failure category	Symptoms	Primary references
Identity churn	Resource addresses shift after refactor, count index churn, missing moved blocks	Code Patterns: count vs for_each, Code Patterns: moved blocks, Code Patterns: LLM mistakes
Secret exposure	Secrets in defaults, state, logs, CI artifacts	Security & Compliance, Code Patterns: write-only, State Management
Blast radius	Oversized stacks, shared prod/non-prod state, unsafe applies	State Management, Module Patterns
Destroy cascade	Targeted destroy deletes more than expected; locals referencing a targeted resource make all for_each consumers implicit dependents	Response Contract: plan-destroy first; State Management: Safe Destroy
CI drift	Local plan ≠ CI plan, apply without reviewed artifact, unpinned versions	CI/CD Workflows, Code Patterns: versions
Compliance gaps	Missing policy stage, no approval model, no evidence retention	Security & Compliance, CI/CD Workflows
Testing blind spots	Plan-only validation of computed values, set-type indexing, mock/real confusion	Testing Frameworks
State corruption / recovery	Stuck lock, backend migration, drift reconciliation	State Management
Provider upgrade risk	Breaking-change provider bump, unpinned modules	Code Patterns: versions, Module Patterns
Provider lifecycle	Removing a provider with resources still in state, orphaned resources, removed block usage	State Management: Provider Removal
Bootstrap / orchestration misuse	null_resource + local-exec for bootstrap, remote-exec for setup scripts, provisioner stdout leaking secrets in CI logs	Code Patterns: Provisioners as Last Resort
Navigation / safe-rename blind spots	Cannot locate symbol defs/refs semantically, value-symbol rename done as blind text replace, grep-only refactor missing refs, hallucinated rg shim	Code Intelligence
Cross-cloud / provider mapping	"What's the Azure/GCP equivalent of X", picking a backend/auth model per cloud	State Management: Cross-cloud equivalents

When to Use This Skill

Activate when: creating or reviewing Terraform/OpenTofu configurations or modules, setting up or debugging tests, structuring multi-environment deployments, implementing IaC CI/CD, choosing module patterns or state organization, configuring or migrating remote state backends.

Don't use for: basic HCL syntax questions Claude already knows, provider API reference (link to docs), cloud-platform questions unrelated to Terraform/OpenTofu.

Core Principles

Module Hierarchy

Type	When to Use	Scope
Resource module	Single logical group of connected resources	VPC + subnets, SG + rules
Infrastructure module	Collection of resource modules for a purpose	Multiple resource modules in one region/account
Composition	Complete infrastructure	Spans multiple regions/accounts

Flow: resource → resource module → infrastructure module → composition.

Directory Layout

environments/   # prod/ staging/ dev/  — per-env configurations
modules/        # networking/ compute/ data/ — reusable modules
examples/       # minimal/ complete/ — docs + integration fixtures

Separate environments from modules. Use examples/ as both documentation and test fixtures. Keep modules small and single-responsibility.

See Module Patterns for architecture principles, naming conventions, variable/output contracts.

Naming Conventions (summary)

• Descriptive resource names (aws_instance.web_server, not aws_instance.main)
• Reserve this for genuine singleton resources only
• Prefix variables with context (vpc_cidr_block, not cidr)
• Standard files: main.tf, variables.tf, outputs.tf, versions.tf

See Module Patterns: Variable Naming and Code Patterns: Block Ordering for examples.

Block Ordering (summary)

Resource blocks: count/for_each first → arguments → tags → depends_on → lifecycle. Variable blocks: description → type → default → validation → nullable → sensitive.

See Code Patterns: Block Ordering & Structure for the full rules and examples.

Testing Strategy

Decision Matrix: Which Testing Approach?

Situation	Approach	Tools	Cost
Quick syntax check	Static analysis	validate, fmt	Free
Pre-commit validation	Static + lint	validate, tflint, trivy, checkov	Free
Terraform 1.6+, simple logic	Native test framework	terraform test	Free-Low
Pre-1.6, or Go expertise	Integration testing	Terratest	Low-Med
Security/compliance focus	Policy as code	OPA, Sentinel	Free
Cost-sensitive workflow	Mock providers (1.7+)	Native tests + mocks	Free
Multi-cloud, complex	Full integration	Terratest + real infra	Med-High

Native Test Rules (1.6+)

Before writing test code: validate resource schemas via Terraform MCP so assertions target real attributes.

• command = plan — fast, for input-derived values only
• command = apply — required for computed values (ARNs, generated names) and set-type nested blocks
• Set-type blocks cannot be indexed with [0] — use for expressions or materialize via command = apply
• Common set types: S3 encryption rules, lifecycle transitions, IAM policy statements

See Testing Frameworks for static-analysis pipelines, native-test patterns, Terratest integration, mock providers, and the full LLM-mistake checklist.

Count vs For_Each — Quick Rule

Scenario	Use	Why
Boolean condition (create / don't)	count = condition ? 1 : 0	Optional singleton toggle
Items may be reordered or removed	for_each = toset(list)	Stable resource addresses
Reference by key	for_each = map	Named access
Multiple named resources	for_each	Better identity stability

Never use list index as long-lived identity — removing a middle element reshuffles every address after it. For the decision matrix, safe migration playbook, moved block patterns, and known-at-plan failure cases, see Code Patterns: count vs for_each.

Locals for Dependency Management

Using try() in a local to prefer a conditional resource's attribute over its parent is a specialized but high-value pattern — it forces correct deletion order without explicit depends_on. Common use: VPC + secondary CIDR associations + subnets.

See Code Patterns: Locals for Dependency Management for the full pattern and worked example.

Module Development

Standard layout:

my-module/
├── README.md       # Usage documentation
├── main.tf         # Primary resources
├── variables.tf    # Typed inputs with descriptions
├── outputs.tf      # Output values
├── versions.tf     # required_version + required_providers
├── examples/
│   ├── minimal/
│   └── complete/
└── tests/
    └── module_test.tftest.hcl   # or Go for Terratest

Variable contracts: always description, always explicit type, use validation for complex constraints, use sensitive = true for secrets, prefer optional() with typed defaults (1.3+) over untyped map(any).

Output contracts: always description, mark sensitive outputs, expose stable subsets (not whole provider objects).

See Module Patterns for the full contract patterns, module release checklist, and LLM-mistake checklist.

CI/CD

Pipeline stages: validate → test → plan → apply (with environment protection).

Cost control: mock providers on PR validation, real-cloud integration only on main or scheduled, tag test resources, auto-cleanup.

Drift prevention: pin runtime and providers, commit .terraform.lock.hcl, apply the reviewed plan artifact from the plan stage (do not re-run plan inside the apply job), run policy/security stage on every path to apply.

See CI/CD Workflows for GitHub Actions, GitLab CI, and Atlantis templates plus the LLM-mistake checklist.

Security & Compliance

Essential checks:

trivy config .
checkov -d .

Don't: store secrets in variables or .tfvars, use default VPC, skip encryption, open security groups to 0.0.0.0/0, use inline ingress/egress blocks in aws_security_group.

Do: source secrets from a cloud secret manager (AWS Secrets Manager / Azure Key Vault / GCP Secret Manager) or use write_only arguments on 1.11+, create dedicated VPCs, enforce encryption at rest and TLS, least-privilege SGs, use separate aws_vpc_security_group_{ingress,egress}_rule resources (e.g. AWS provider v5+).

Marking a variable sensitive = true masks display only — the value still lives in state. Use write_only / *_wo on 1.11+, or keep secret material out of Terraform entirely via runtime lookups.

See Security & Compliance for trivy/checkov pipelines, state-file hardening, compliance mappings, and the LLM-mistake checklist.

State Management

Never use local state in teams or production. Remote backends provide automatic locking, encryption, versioning, audit logging, and safe collaboration.

Choosing a Remote Backend

AWS example (Azure azurerm / GCP gcs / TF Cloud syntax: see State Management: Choosing a Remote Backend):

terraform {
  backend "s3" {
    bucket        = "my-terraform-state"
    key           = "prod/vpc/terraform.tfstate"
    region        = "us-east-1"
    encrypt       = true
    use_lockfile  = true   # Native S3 locking, 1.10+
  }
}

On Terraform < 1.10, use dynamodb_table = "terraform-state-lock" instead of use_lockfile. Azure Storage, GCS, and Terraform Cloud all offer built-in locking - see the State Management reference for syntax. For choosing among backends and their locking models, see Choosing a Remote Backend.

State Organization

Pattern	Use When	Example Path
Per environment	Different teams per env	prod/terraform.tfstate, staging/...
Per component	Independent lifecycles	prod/vpc/, prod/eks/, prod/rds/
Hybrid (recommended)	Both benefits	prod/networking/, prod/compute/, staging/networking/

Split state when: different teams, different update cadences, or >500 resources. Combine when: tightly coupled resources, <100 resources, same lifecycle.

See State Management for locking, migration, multi-team isolation, disaster recovery, and the LLM-mistake checklist.

Version Management

Component	Strategy	Example
Terraform runtime	Pin minor	required_version = "~> 1.9"
Providers	Pin major	version = "~> 5.0"
Modules (prod)	Pin exact	version = "5.1.2"
Modules (dev)	Allow patch	version = "~> 5.1"

Commit .terraform.lock.hcl intentionally. Keep provider/runtime upgrades in a separate PR from functional changes. See Code Patterns: Version Management for constraint syntax and upgrade workflow.

Modern Terraform Features (1.0+)

Feature	Min version	Common use
try()	0.13+	Safe fallbacks, replaces element(concat())
nullable = false	1.1+	Prevent null silently overriding defaults
moved blocks	1.1+	Refactor without destroy/recreate
optional() with defaults	1.3+	Typed object attributes
import blocks	1.5+	Declarative imports, reviewable in VCS
check blocks	1.5+	Runtime assertions
Native terraform test	1.6+	Built-in test framework
Mock providers	1.7+	Cost-free unit testing
removed blocks	1.7+	Declarative resource removal
Provider-defined functions	1.8+	Provider-specific transformations (requires provider to declare functions)
Cross-variable validation	1.9+	Reference other var.* in validation blocks
write_only arguments	1.11+	Secrets never stored in state
S3 native lock-file	1.10+	State locking without DynamoDB

Before emitting a feature, verify the runtime floor. See Code Patterns: Feature Guard Table for the full table with common LLM error patterns per feature.

Runtime-Specific Guidance

• Terraform 1.0-1.5 (OpenTofu starts at 1.6): Terratest for integration, static analysis + plan validation only (no native tests).
• 1.6+: native terraform test / tofu test available — migrate simple unit tests, keep Terratest for complex integration.
• 1.7+: mock providers cut test cost — mock for unit tests, real runs for final integration.
• 1.10+: S3 native lock-file (use_lockfile) is the correct default for new configurations — DynamoDB locking is no longer required.
• 1.11+: write_only arguments for secret handling keep credentials out of state.
• Terraform vs OpenTofu: both supported. For licensing, governance, and feature delta, see Quick Reference: Terraform vs OpenTofu.

Code Intelligence (terraform-ls)

Semantic navigation for HCL. terraform-ls is optional; without it every row below degrades to a disclosed rg + Read fallback.

Self-contained terraform-ls layer of a generic code-intelligence discipline - apply the rows below directly. Recommended companion: the code-intelligence plugin (same antonbabenko/agent-plugins marketplace) carries the generic discipline (position anchoring, degradation gate, disclosure format, anti-phantom-shim) and ships /code-intelligence:doctor for readiness. If it is installed, defer to its generic protocol; this skill stays fully self-contained without it.

Goal	Use	Tradeoff
Find definition / all references	terraform-ls goToDefinition / findReferences	Needs init + a position anchor
Rename value symbol (var/local/output/provider alias)	Manual: findReferences -> per-file fresh Read -> edit -> validate	No rename provider
Rename resource/module address	moved block + plan shows 0 destroy	Text rename forces destroy/recreate
Exact text / known name / .tfvars / non-HCL	rg + Read	No semantic scope

✅ Supported: goToDefinition, findReferences, documentSymbol, hover, workspaceSymbol. ❌ Unsupported: goToImplementation, call hierarchy, rename provider. Do not call these then report their absence as a finding.

• ✅ Prereq: local terraform/tofu on PATH, terraform init run; cold start may need one retry.
• ✅ LSP calls are position-anchored (file:line:character) - anchor with rg first, never symbol-name-only.
• ❌ Do not claim "LSP broken, using rg" until the Degradation Gate passes; disclose any tool substitution on the first line.

Depth: Code Intelligence.

Reference Files

Progressive disclosure — essentials here, depth on demand:

• Testing Frameworks — static analysis, native tests, Terratest, mock providers
• Module Patterns — structure, variable/output contracts, terraform_remote_state rules, release checklist
• CI/CD Workflows — GitHub Actions, GitLab CI, Atlantis, cost control
• Security & Compliance — trivy/checkov, secrets handling, compliance mappings
• State Management — backends, locking, migration, multi-team, recovery
• Code Patterns — block ordering, count/for_each deep dive, modern features, version management, locals
• Code Intelligence - terraform-ls capabilities, position-anchored calls, manual rename, degradation gate
• Quick Reference — command cheat sheets, flowcharts, troubleshooting

License

Apache License 2.0. See LICENSE for full terms.

Copyright © 2026 Anton Babenko