review>

Stop Review Fatigue!

Review Raven puts easy-to-digest design and code reviews right in your agentic loop.

It's a command-line tool + agent skill that presents a structured summary of the changes when the agent is done coding. The agent prepares the review itself and explains the change so it is easy to digest for humans. Review Raven checks it against Git to keep the agent honest.

  • Which parts of the system did the agent touch?
  • Which decisions did it make without consulting you?
  • Is the design ideal and if not, which caveats do you need to be aware of?

Plus a walkthrough of the diff in a logical order.

If you find anything you don't like, you comment on any of these elements. When you submit, the agent addresses all of your comments and starts another Review Raven round.

:try the demo :install

Your agent writes code faster than you can review it.

A diff shows what changed. It does not show which alternatives the agent rejected, which assumption it made without asking, or where it is unsure.

So you skim, approve, and find out later. That is review fatigue, and it is how vibe coding ends up in production.

Review Raven puts you back in the driver's seat: the agent explains itself, and you decide.

Use cases

How it works

  1. 01

    The agent explains

    When it finishes a change, the agent writes a review of it: architecture, design, decisions, caveats, walkthrough.

  2. 02

    We check it against git

    Every diff is resolved from git. Changed files the review leaves out are flagged. Nothing is invented.

  3. 03

    You comment

    Comment on lines, decisions, the design verdict or caveats, in a local page.

  4. 04

    The agent reacts

    Approve the change if you like it. Submit comments, and it answers each one and serves the next review round.

What a review shows

Architecture

The concepts the change touched, the system around them, and how they relate.

Ideal design

Would the agent build it the same way on a clean slate? If not, it says what it would build.

Decisions

Every choice made without you: the constraint behind it, its cost, and what it beat.

Caveats

What will surprise you, stated before it does.

Diff narrative

The diff, one concept at a time, with a passage on why each chunk exists.

Checked against git

No omissions, no hallucinated diffs. A changed file the review skips is flagged.

Comment on anything

Lines, decisions, design, caveats. The agent answers each comment in the next round.

Asked, not guessed.

Pull-request bots have a second agent reconstruct intent from the diff, after it is pushed. Review Raven asks the agent that made the decisions, while you can still change them.

Built-in diff views
Accept or reject hunks. Little explanation.
PR review bots
Summaries inferred after the fact, on pushed code, by an agent that did not write it.
Review Raven
The authoring agent states its design verdict, decisions and caveats, checked against git, before you commit.

Install

A plugin for Claude Code and GitHub Copilot CLI. No toolchain; the binary for your platform is fetched on first use.

:install guide :try the demo