Terraform Engineer gives the agent a working method for infrastructure as code. It analyzes the existing setup, designs reusable modules, configures remote state with locking and encryption, and runs format, validate, lint and test checks before it plans.
It ends with a plan summary that highlights creates, updates and destructive changes. The agent applies only after you confirm, and it refuses to apply destructive changes you have not accepted.
What You Get
- A seven-step workflow from analysis to an approved apply
- Module structure with typed, validated variables, described outputs and the standard file layout
- Remote backend configuration with native S3 locking and encryption
- Recovery steps for validation failures, state drift, provider auth errors and ordering problems
- Five reference files for modules, state, providers, testing and best practices
Common Use Cases
- Writing a reusable module with input validation and versioning
- Migrating a backend or importing existing resources into state
- Adding native terraform test files for a module
- Setting up provider pinning and a lock file for several clouds
What Makes It Unique
The skill puts an approval gate into the workflow. The agent must present a plan summary and wait for your confirmation, and any command that writes state (refresh-only apply, import, state mv, state rm) needs approval first. It also names deprecated constructs to avoid in new code, such as terraform refresh, DynamoDB state locking and tfsec, and points to current replacements like use_lockfile, moved and import blocks, and Trivy.
Under The Hood
The skill loads SKILL.md and pulls in one reference file at a time when the task matches its topic. It tells the agent to run shell commands: terraform fmt, validate, plan, test and apply, plus tflint and trivy or checkov when installed. The plan is written to a plan file on disk and applied only after you approve it. The commands read cloud credentials and provider settings from your environment, can change real infrastructure on apply, and can modify state files. The skill tells the agent never to print state, because state can hold sensitive values. The skill itself contains no scripts.