buddhikernel

Buddhi Review — AI skill for Claude Code

AI community

Automated multi-reviewer pull-request review for Claude Code: fan out AI reviewers, classify findings, guide fixes, and merge clean.

How to install Buddhi Review

This entry records only its repository, not the path inside it, so there is no exact command to give. Open buddhikernel/buddhi-review and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Buddhi Review does

Automated multi-reviewer pull-request review for Claude Code: fan out AI reviewers, classify findings, guide fixes, and merge clean.

Alternatives in AI

Codex Plugin — 's /codex:adversarial-review command 9.6k ★
Mancode — AI coding agent harness 358 ★
AI Pair Collaboration — Coordinate heterogeneous AI teams: one creates, two review from different angles 190 ★

README

buddhi-review

[![PyPI version](https://img.shields.io/pypi/v/buddhi-review.svg)](https://pypi.org/project/buddhi-review/) [![Python versions](https://img.shields.io/pypi/pyversions/buddhi-review.svg)](https://pypi.org/project/buddhi-review/) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/buddhikernel/buddhi-review/blob/main/LICENSE)

`buddhi-review` is a **free, MIT-licensed PR review-and-fix loop for Claude Code**, built on the public [Buddhi kernel](https://github.com/buddhikernel/buddhi). It sends each PR to a cross-vendor panel of AI reviewers, classifies their findings, applies fixes, and repeats until the pull request is ready to land. It auto-merges only when you opt in.

**The author-model, the model used to produce the code, should not be the only reviewer.** Research documents **self-preference bias**, in which an LLM evaluator can favor output from the same model. A single review pass can also leave many bugs undiscovered. That is why Buddhi uses a [cross-vendor panel and re-reviews the code after each round of fixes](#why-a-panel-and-why-rounds).

Across [88 qualifying internal runs](#what-real-review-runs-show) on a codebase written with Claude Code, the four-reviewer panel of Claude, Codex, Gemini, and Copilot found 681 valid bugs. Claude, the author-model in these runs, found 3.8% of them; reviewers from other vendors found 96.2%, and half of the valid bugs surfaced only in round 2 or later. This result measures Claude in the author-model role on this codebase; it is not a general ranking of Claude's code-review ability.

A Buddhi review loop: five rounds of cross-vendor review, fixes, and re-review before the PR lands.