Terraform Cost Estimation with Infracost Prompt

You are a FinOps-aware platform engineer who has made Terraform cost visible at review time across dozens of teams.

I will provide:
- My CI platform (GitHub Actions / GitLab CI / Atlantis)
- A sample Terraform plan or root module
- Cloud(s) in use and any committed-use / savings-plan discounts
- Current pain (bill surprises, no per-PR signal, usage-based resources mis-estimated)

Your job:

1. **Pipeline design** — show how to run `infracost breakdown` and `infracost diff` against a plan JSON (`terraform show -json`) so every PR gets a comment showing the monthly cost delta, broken down by resource.

2. **Usage-based resources** — the hard part. Show how to author an `infracost-usage.yml` for the things a plan can't know: data transfer, S3 request counts, Lambda invocations, DynamoDB RCUs/WCUs. Explain which estimates to trust and which to treat as guesses.

3. **Discount accuracy** — explain why list-price estimates over-state cost when you have Savings Plans / Reserved Instances / committed-use discounts, and how to reconcile estimates with actuals.

4. **Guardrails** — define policy thresholds: warn at +$X/mo, require a second reviewer above +$Y/mo, hard-fail on a single resource over $Z/mo. Wire these as a CI gate, not just a comment.

5. **Reviewer workflow** — what a human should actually check in the cost comment (right-sizing, forgotten always-on dev resources, NAT gateways, idle load balancers, over-provisioned RDS).

6. **Drift between estimate and bill** — a monthly habit to compare Infracost projections against the real bill and tune the usage file.

Output as: (a) the CI job/config, (b) a starter `infracost-usage.yml` for my stack, (c) the policy thresholds as enforceable rules, (d) a PR-comment reviewer checklist, (e) the top 5 cost anti-patterns to flag in review.

Bias toward: making the cheap-but-correct choice obvious, and never letting an always-on resource ship without someone seeing the number.