A field manual for Terraform, aimed at the agent holding the keyboard

What it does

It’s an Agent Skill: a curated body of Terraform and OpenTofu best practices that an AI coding agent loads on demand instead of improvising. When you ask for a module with tests, a remote state setup, or a GitHub Actions pipeline, the agent draws on decision matrices, DO/DON’T patterns, and version-specific guidance (native tests, use_lockfile, write_only) rather than its training-data hunch. AWS is the default dialect, but Azure and GCP are treated as first-class — ask for the equivalent pattern and the skill maps it.

The interesting bit

The content isn’t scraped from a wiki; it’s distilled from terraform-best-practices.com, the terraform-aws-modules collection, and the author’s enterprise IaC experience. The pitch is decision frameworks — not just “do this” but when and why — which is exactly the judgment agents lack.

Key highlights

• Testing guidance: native terraform test vs Terratest, with static/integration/E2E workflows
• State management: remote backends, locking, multi-team isolation, migration and recovery
• CI/CD: GitHub Actions, GitLab CI, Atlantis, Infracost cost estimation, Trivy/Checkov scanning
• Works across Claude Code, Cursor, Copilot, Gemini CLI, Codex, Kiro, and any Agent Skills-compatible host
• Optional terraform-ls integration; falls back to plain text search without it

Caveats

• It’s guidance, not code — the value depends on your agent actually loading and following it.
• Some version-specific features (use_lockfile 1.10+, write_only 1.11+) assume a recent Terraform or OpenTofu.
• The README warns of a marketplace-name clash if you add this repo directly alongside antonbabenko/agent-plugins — install via one or the other.

Verdict

If an agent writes your Terraform, this is a cheap way to make it write Terraform like someone who has run it in production. If you don’t use AI coding agents, there’s nothing here for you.