← ClaudeAtlas

planguard-terraformlisted

Reviews a Terraform plan before it is applied: what gets destroyed or replaced, what gets exposed to the internet, what loses its guardrails. Runs tf_plan_review.py on the JSON from `terraform show -json` (20 rules, sensitive values masked), ranks the findings, explains each one in plain language and proposes the smallest safe change. Use when the user says "review my terraform plan", "is this plan safe to apply", "what will terraform destroy", "check this plan.json", "why is terraform replacing my database", or "can I apply this".
vivekbhange-hub/plan-guard-skill · ★ 0 · DevOps & Infrastructure · score 75
Install: claude install-skill vivekbhange-hub/plan-guard-skill
# planguard-terraform: read the plan before the plan reads your production `terraform plan` prints a wall of text that people scroll past. The dangerous lines are few: a database that says `must be replaced`, a security group that gained `0.0.0.0/0`, an IAM policy that became `"*"`. This skill runs `tf_plan_review.py` over the machine-readable plan, then does the part a script cannot: explain each finding to this user and write the smallest change that removes it. ## Inputs - The plan as JSON. The user produces it themselves, in the Terraform folder: ``` terraform plan -out=plan.out terraform show -json plan.out > plan.json ``` Claude does not run these. `plan.json` can contain secrets (passwords, tokens, keys in resource arguments), so keep it out of git and never paste it into chat. The script reads the file by path and masks every value the plan marks sensitive. - Optional: a `.planguard-ignore` file with `RULE ADDRESS-GLOB reason` lines for accepted findings. - Commands below say `python3`; on Windows use `python` (or `py -3`). ## Outputs - A report: summary counts (add, change, destroy, replace), a verdict, every finding with rule id, severity, what happens, why it matters and the safe change, and a "cannot be reviewed" list. - For each finding you keep, a concrete patch to the `.tf` code (a lifecycle block, a `moved` block, a tightened CIDR, a scoped policy) and what to re-check after it. - A short ranked summary for the human: the three things to