Skip to content

Expert's corner

AI code review on every pull request with Claude Code and GitHub Actions

A code change with three AI review comments, marked red, orange and blue by severity

Evenbytes has automated AI code review in the development of a leading process digitalisation platform. Whenever someone opens or updates a pull request (PR), a GitHub Actions workflow runs a review with Claude. The review posts its findings as a comment on the PR itself, ordered by severity, before anyone approves it.

Automated review is one part of a wider change in the way the team develops: the PRD + TDD with Claude Code methodology.

How it works

The review is built into the repository’s CI/CD and runs automatically on every PR:

  1. When a pull request is created or updated, GitHub Actions runs the review workflow.
  2. A Claude Code skill defines what to review and how. The review is carried out by Claude, split into subagents with scoped rules: each one focuses on one type of problem.
  3. A verification pass checks the findings before they are posted, so that false positives do not get through.
  4. The bot posts a comment on the PR with the classified findings.

AI code review flow: GitHub Actions runs Claude subagents, the findings are verified and posted on the pull request and, after the CI tests, Cloud Build deploys or blocks the change

Blockers, warnings and suggestions

The review comment sorts the findings into three levels:

  • Blockers: problems that must be fixed before the change is merged.
  • Warnings: risks worth checking, such as edge cases in status logic, null dates or mismatches with the business rules.
  • Suggestions: optional improvements.

Illustrative example of an AI review comment on a pull request, with a blocker, a warning and a suggestion

This way, whoever reviews the PR starts with what matters, and the most common mistakes are caught before they reach a person.

Part of a methodology: PRD + TDD with Claude Code

Automated review closes a cycle that starts before any code is written:

  • PRD: every piece of development starts from a requirements document.
  • TDD: the tests are written first, then the code that passes them, with Claude Code as an assistant.
  • AI review: every PR goes through the automated review before it is merged.

The goal is a pipeline that does not depend on manual approvals. Unit and automated tests run in CI, and clear rules decide which changes are deployed to Google Cloud with Cloud Build and which are blocked.

Keeping an eye on costs

Each review uses GitHub Actions minutes. That is why the team monitors the minutes each developer generates when opening or updating PRs, so that automation does not drive up spending.

What comes next

If your team wants to bring AI into its development process, from requirements to review and deployment, take a look at our software development and process automation services, or talk to an Evenbytes expert.