iac
Routes Infrastructure as Code requests to the correct technology skill and compares Terraform, OpenTofu, Pulumi, CloudFormation, Bicep, and Ansible for provisioning strategy. WHEN: "infrastructure as
SKILL.mdplugins/devops/skills/iac/SKILL.md
GitHubInfrastructure as Code Router
This skill routes Infrastructure as Code requests to the right technology skill and covers cross-tool comparison. Determine which technology best matches the request, then read that skill's SKILL.md for implementation depth.
Decision Matrix
| Signal | Skill |
|---|---|
| Terraform, HCL, providers, modules, state, workspaces, plan/apply | terraform |
| OpenTofu, tofu plan, tofu apply, Terraform fork, MPL license | opentofu |
| Pulumi, pulumi up, TypeScript/Python/Go IaC, ComponentResource | pulumi |
| CloudFormation, CFN, AWS stacks, StackSets, SAM, CDK | cloudformation |
| Bicep, ARM template, Azure Resource Manager, az deployment | bicep |
| Ansible, playbooks, roles, inventory, modules, AWX, Tower, Jinja2 | ansible |
| IaC comparison, "which tool", provisioning vs config management | Handle directly (below) |
How to Route
- Extract technology signals from the user's question — tool names, file extensions (.tf, .yml), CLI commands (terraform, ansible-playbook), provider names.
- Check for version specifics — if a version is mentioned (Terraform 1.15, Ansible 2.20), read the matched technology skill; version-specific nuance lives in its
references/versions/directory. - Comparison requests — if the user is comparing IaC tools, handle directly using the framework below.
- Ambiguous requests — if the user says "automate my infrastructure" without specifying a tool, gather context (cloud provider, existing tooling, team skills) before routing.
IaC Fundamentals
Load references/concepts.md when the user needs foundational understanding of IaC patterns that apply across all tools.
Tool Selection Framework
Provisioning vs Configuration Management
This is the most important distinction in IaC:
| Category | Purpose | Tools | Analogy |
|---|---|---|---|
| Provisioning | Create/destroy infrastructure resources (VMs, networks, databases, DNS) | Terraform, Pulumi, CloudFormation, Bicep | Building the house |
| Configuration Management | Configure existing servers (install packages, manage files, start services) | Ansible, Chef, Puppet, SaltStack | Furnishing the house |
Terraform and Ansible are complementary, not competing. Terraform creates the VM; Ansible configures it. The overlap is in simple provisioning (Ansible can create cloud resources) and simple configuration (Terraform provisioners can run scripts).
When to Use Each Tool
| Scenario | Best Tool | Why |
|---|---|---|
| Multi-cloud infrastructure provisioning | Terraform | Provider ecosystem, state management, plan/apply workflow |
| AWS-only infrastructure | Terraform or CloudFormation | CloudFormation has deeper AWS integration; Terraform is more portable |
| Azure-only infrastructure | Terraform or Bicep | Bicep has native Azure support; Terraform for multi-cloud |
| Server configuration at scale | Ansible | Agentless, SSH-based, idempotent modules, wide OS support |
| Immutable image builds (AMIs, VM images) | Packer + Ansible | Packer orchestrates; Ansible provisions the image |
| Developers who want real programming languages | Pulumi | TypeScript, Python, Go, C# instead of DSLs |
| Kubernetes resource management | Helm, Kustomize, or ArgoCD/Flux | Purpose-built for K8s; Terraform K8s provider is awkward |
| Network device configuration | Ansible | Network modules for Cisco, Juniper, Arista, Palo Alto |
Comparison Matrix
| Dimension | Terraform | Ansible | Pulumi |
|---|---|---|---|
| Model | Declarative | Procedural (idempotent tasks) | Declarative (imperative syntax) |
| Language | HCL | YAML + Jinja2 | Python, TypeScript, Go, C# |
| State | Remote state file (required) | Stateless (agentless) | Managed or self-hosted state |
| Agent | None | None (SSH/WinRM) | None |
| Strengths | Plan preview, provider ecosystem, modules | Agentless, simple, config management | Real languages, IDE support, testing |
| Weaknesses | State complexity, HCL limits | Slow at scale, ordering matters | Vendor risk, debugging complexity |
| Community | Massive (largest IaC ecosystem) | Massive (most popular config mgmt) | Growing (smaller than Terraform) |
| Licensing | BSL 1.1 (HashiCorp) | GPL v3 (Red Hat) | Apache 2.0 (core) |
Anti-Patterns
- "Terraform for configuration management" — Using Terraform provisioners (remote-exec, local-exec) for server config. Use Ansible or cloud-init instead.
- "Ansible for infrastructure provisioning at scale" — Ansible can create cloud resources, but without state tracking it can't detect drift or plan changes. Use Terraform for provisioning.
- "One giant Terraform root module" — Monolithic state files are slow, risky, and hard to maintain. Decompose into smaller, independent state units.
- "No remote state" — Local Terraform state in a team = guaranteed conflicts and data loss.
- "Click-ops then import" — Creating resources manually then importing into Terraform. Start with IaC from day one.
Reference Files
references/concepts.md— IaC principles (state management, idempotency, immutable infrastructure, drift detection). Read for foundational or comparison questions.