DEV Community

Anur Bašić
Anur Bašić

Posted on Edited on

Reflektor - Code review on steroids

Check it out on GitHub: github.com/bashovski/reflektor

A little rant, and why I built Reflektor

Microsoft went down the other day, and it took them a long time to get back up. The question is, do they even have any developers at the firm who actually know how things work?

Now it's just prompting LLMs and praying that they'll find the issue.

Code gets pushed blindly. We lowered the bar so we can ship faster.

You used to write every single line of code, and you knew what it did. Now, you have no clue.

Devs ask other devs to review their PRs without looking at a single line of code themselves. Guilty yet?

This is nothing new, but how does one protect their codebase from AI slop and actually have some ownership?

I held onto an idea of a high-level, quality PR report for a long time, until I made it this week and released it.

Reflektor.

It's built with short review times in mind and oriented towards the direction a PR is taking, rather than focusing on variable names.

While thinking of where reviewers would save the most time, a few things came to mind:

  • Being able to see the actual UI that's new or has been changed
  • Knowing what to review and what to skim
  • Being aware of the Database model/schema changes
  • Looking for new types, interfaces, etc.
  • Knowing any risks before a PR goes live

Do I need to know every single line of code?

  • No. Do I need to know whether your PR ships 1,000 lines of slop, or is it a meaningful PR? - Yes!

What's released, and what's to come

In the initial release that went live just this morning, a skill that you can run yourself in any codebase is fully ready and tested on all major agents, and it works.

The skill is tech-agnostic, meaning it's working with any tech stack you throw its way. When you invoke/run a skill, it runs a daemon in the background and it additionally augments your reports and strengthens your Developer Experience, while everything runs locally.

Optionally, you can connect to reflektor.dev and push reports automatically to the platform and have your teammates review them.

Feel free to run as many verification/review loops, because that's what is gonna turn any PR into a good PR.

Get Started

To get started, I'd strongly advise you to open the GitHub Repo or the Install Page and run the skill on one of your sessions.

You can simply install the skill:

npx skills add bashovski/reflektor#skill -g
Enter fullscreen mode Exit fullscreen mode

and then run:

/reflektor
Enter fullscreen mode Exit fullscreen mode

It's the best way to get started, and you'd save time having to read every single line of code. AI doesn't optimize code generation for readability by default. So, additionally, you may want to tweak your AGENTS.md or CLAUDE.md further to do so.

List of features

  • Technology-agnostic
  • Review Loops
  • Collaboration
  • Automatic Project Checks
  • E2E UI Testing with Screenshots
  • Flow Visualizations
  • Database Changes Visualizations
  • Test Coverage Summaries and Reports
  • Hooks (allows pipelines customization)
  • Daemon (runs in background)

Top comments (0)