Vale construir sua versão de Infracost Cloud?

Build a self-hosted GitHub App called CostLens in an empty repository.
Use Python 3.12, FastAPI, SQLite, python-hcl2, httpx, PyJWT, and Docker Compose; do not offer alternative stacks.
The core loop is: when a pull request changes Terraform, estimate the AWS monthly cost delta, run basic FinOps checks, and update one PR comment.
Authenticate as a GitHub App using an app ID, installation token flow, webhook secret, and private key stored outside the repository.
Handle pull_request opened, synchronize, and reopened events, and verify every webhook signature before processing.
Analyze the base commit and head commit separately so the comment shows previous cost, proposed cost, and monthly delta.
Parse Terraform HCL locally and support a deliberately bounded first set: EC2 Linux on-demand instances, EBS gp2/gp3 volumes, and RDS on-demand instances plus allocated storage.
Use the public AWS Price List Bulk files as the pricing source, cache downloaded price data, and record the source publication timestamp.
Assume 730 hours per month for hourly resources and show every assumption in the result.
Never price an unsupported or ambiguous resource as zero; list it as unsupported and exclude it from totals.
Support count when it resolves to a literal integer; mark for_each, unknown expressions, modules, and unresolved variables as unsupported in v1.
Add configurable required-tag checks and a configurable monthly cost-increase warning threshold.
Post a Markdown PR comment with total before, total after, delta, the five largest increases, policy warnings, unsupported resources, and pricing timestamp.
Use a hidden HTML marker so repeated webhook deliveries update the existing comment instead of creating duplicates.
Store repository settings and analysis history in SQLite, with a small admin page for recent PR analyses and configuration.
Add retry and timeout handling for GitHub and AWS pricing downloads, structured logs, and a health endpoint.
Ship .env.example, a Dockerfile, docker-compose.yml, database migrations, and one-command local startup.
Include unit tests for pricing normalization and Terraform resource extraction plus one end-to-end webhook happy path using fixtures.
Document the minimum GitHub App permissions and events in the README, including pull requests read, contents read, and issues write.
Do not send Terraform source, secrets, or repository contents to any AI or third-party analysis API.
Deliberately leave out Azure, GCP, Terragrunt, CloudFormation, CDK, negotiated discounts, Savings Plans, Reserved Instances, and usage-based cost forecasting.
Deliberately leave out SSO, billing, organization dashboards, campaigns, AutoFix, enterprise RBAC, and GitLab or Azure DevOps integrations.
Finish by running the tests, starting the stack locally, processing the included sample webhook, and listing the exact commands used.