Agent Review

agent-review.el

Overview
https://github.com/user-attachments/assets/8ad86c2d-a91e-4f69-b51a-0f7f72ae0b19

agent-review enables a streamlined workflow: use one AI agent (Claude, Cursor, Gemini) to implement features, then instantly get a second opinion from a different agent. The package automatically collects your git changes and sends them to your chosen AI agent for review. The cross-agent review catches issues that single-agent workflows miss, while the tight feedback loop means you fix problems quicker and more efficiently.

All findings display in a navigable list interface where you can jump to issues, triage by severity, and send fixes back to your implementation agent—closing the loop without leaving Emacs.

Requirements

- Emacs 29.1 or later
- acp.el >= 0.7.1
- agent-shell >= 0.16.2
- Git
- An ACP-compatible agent (Claude Code, Cursor, Gemini CLI, etc.)

This package builds on the excellent work by xenodium on acp.el and agent-shell. Consider supporting their work!

For setting up individual agents (Claude Code, Cursor, Gemini CLI, etc.), follow the agent-shell setup guide.

Installation

Using straight.el

Using Doom Emacs

Add to your packages.el:

Manual

Clone this repository and add to your load path:

Usage

Basic Usage

1. Make some changes in a git repository
2. Run M-x agent-review
3. Select an agent from the list
4. Wait for the review to complete
5. Browse issues in the *Agent Review* buffer

Key Bindings (in review buffer)

Programmatic Usage

Configuration

Default Agent

Set a preferred agent to skip the selection prompt:

Git Executable

If git is not in your PATH:

How It Works

1. Collection: Runs git diff and git diff --cached to get changes
2. Analysis: Sends changes to AI agent with structured prompt
3. Parsing: Extracts issues in format…