tf-plan-review

Analyze Terraform plans for risk before you apply. Classifies every change as safe, moderate, dangerous, or critical. Detects destroys, IAM changes, data-los...

Before installing or running this skill: - Expect to have jq and terraform (or tofu) on PATH — the registry metadata incorrectly lists no required binaries. Confirm these are present. - Review the included scripts yourself (scripts/tf-plan-review.sh) before trusting it in production. The script is mostly conservative, but it does run `terraform init` if the directory is not initialized, which creates/changes .terraform and can download providers/modules from the network. - If you are concerned about downloading untrusted providers, run the tool in a sandbox or on a copy of the repo that has already been initialized, or set TF_BINARY to a vetted wrapper. - The script writes the plan JSON to a secure temporary file and removes it on exit, but temporary files can persist until deleted; avoid running it on hosts where tmp is not trusted or where disk-forensics is a concern. - Do not grant this skill more credentials than necessary. Because it relies on Terraform, it will use whatever provider credentials Terraform finds; run it with least privilege or in a read-only/staging account if possible. - The discrepancies (metadata omitting dependencies, SKILL.md claiming purely in-memory while script uses a temp file) justify manual review; if you cannot review the script, treat the skill as potentially risky and test it in isolated environments first.

Type: OpenClaw Skill Name: tf-plan-review Version: 0.2.1 The skill is designed as a read-only Terraform plan analyzer. The `SKILL.md` explicitly instructs the AI agent with 'CRITICAL RULES' to NEVER run `terraform apply`, `destroy`, or any state-modifying commands, and to handle sensitive data carefully. The `scripts/tf-plan-review.sh` code adheres to these rules, using `set -euo pipefail`, consistently quoting paths to prevent shell injection, employing `jq --arg` for safe JSON interpolation, and running `terraform plan -out=/dev/null` to avoid writing plan files to disk. The `SECURITY.md` further reinforces these safety measures. There is no evidence of intentional malicious behavior, data exfiltration, or backdoors. The requested `exec` and `network` permissions are justified for its stated purpose of running Terraform commands and interacting with cloud provider APIs.