Triage after automated review
A new open-source developer tool called Farol aims to help human approvers decide where to concentrate attention in a pull request. Distributed as both a command-line package and a GitHub Action, it analyzes a change with static program information and produces a report identifying risky files, affected application surfaces, gaps in diff coverage and possible contract breaks.
The project is designed to follow an AI or conventional line-by-line review, not replace one. Its premise is that an automated reviewer may comment on individual edits while still missing the operational reach of a small change. Farol therefore focuses on the merge decision: which pages, endpoints, jobs or shared components could be affected, and what should be tested manually before approval.
Farol’s repository says its core scoring is deterministic because it relies on compilers and parsers rather than a model’s judgment. The same pull request and configuration should consequently produce the same result. Reports can be generated locally or in continuous integration, and the GitHub Action can comment on pull requests without blocking a merge.
Multiple stacks and evidence sources
The tool documents support for TypeScript and React ecosystems, Go services and Python applications. On the web side it can map React Router, Next.js and TanStack Router pages, along with GraphQL operations and common REST frameworks. Go analysis covers several routers, gRPC services and binaries. Python coverage includes FastAPI, Flask, Django, MCP tools, Celery tasks and workflow systems.
A project configuration file tells Farol about the stack and relevant directories. Teams can optionally feed it test coverage, lint and dependency-audit reports. The generated view then combines changed-file risk with route reach and testing information, giving reviewers a starting order instead of treating every line as equally important. Infrastructure and continuous-integration files are called out separately.
The repository emphasizes that Farol parses code without executing it and does not send source to a hosted service. It runs on a developer machine or CI runner and is available under the MIT license. The package requires Node.js 22 or newer and Git, with language tools needed when analyzing Go or Python repositories. Users can invoke it through npx or install a global command.
Optional Datadog integration can add traffic and error context to a report. The project says this information does not change the score; it instead shows production usage, linked errors and service callers. That integration requires the team to configure its own observability credentials in CI, whereas the core static analysis operates without an account.
Farol reflects a broader shift in code review as automated systems produce and inspect more changes. Its contribution is a risk map for the person who remains responsible for approval. The repository provides a live demonstration using a full-stack FastAPI and React example, but the supplied evidence does not include independent tests of its precision. Teams adopting it should compare its reach maps and warnings against their architecture before relying on the suggested review order.
