What problem does it solve? Reviewing Terraform/OpenTofu infrastructure code for security flaws, state-safety risks, and module-contract issues is slow and error-prone when done manually, and a single reviewer often misses failure modes that another model or scanner would catch. ## Core Features & Use Cases - Multi-agent review: Dispatches the review to one or more configurable LLMs (claude, codex, gemini) as independent subagents, then merges and cross-validates their findings. - Failure-mode rubric: Checks identity churn, secret exposure, blast radius, state safety, module contracts, and testing gaps with a version-aware guard for the detected Terraform/OpenTofu version. - Consent-gated execution: Runs a metadata-only dry-run preview first, excludes .tfvars by default, and requires explicit approval before dispatching file contents to providers, running host scanners (tflint, trivy, checkov), or posting to a PR. - Use Case: Point it at an infra/ directory before merging; it collects the in-scope .tf files, reviews them with codex and gemini in parallel, dedups findings, and exits non-zero if critical or high issues remain so it can gate a pipeline. ## Quick Start Ask the assistant to review the Terraform code in the current directory with stark-terraform-review, approve the file list shown in the dry-run preview, and confirm dispatch to the configured LLM agents.
