slop-review

An agent asks for a code review; a human gives one in a native window; the agent gets back the annotations and nothing else.

A remark being written against a line of the diff. — A remark against one line. Only these travel back to the agent.

A misspelled word underlined in a remark, with suggestions offered. — Checked against your own dictionaries, through your own enchant.

Download

AppImage · x86_64

One file. Make it executable and run it — no install, no runtime, no webview. It carries its own difftastic; git, jj and your dictionaries stay yours. Needs glibc 2.39 or newer, which is Ubuntu 24.04, Debian 13 and anything since.

chmod +x slop-review-x86_64.AppImage
./slop-review-x86_64.AppImage

With nix

Try it without installing anything:

nix run --accept-flake-config gitlab:c0va23/slop-review/stable

Installation has the NixOS and home-manager modules, the binary cache and what to do when it compiles anyway.

For the agent

What an agent needs to know — the call, the exit status, the shape of what comes back — is one skill file in the Agent Skills format, which Claude Code, OpenCode and most other coding agents read. Claude Code installs it as a plugin, straight from the repository:

claude plugin marketplace add https://gitlab.com/c0va23/slop-review.git
claude plugin install slop-review@slop-review

Teaching the agent has the other agents, and what the skill promises them.

Manual

Installationwith nix, or as one file
Usethe command line; reading, remarking and ticking off in the window
Keeping your placewhat is remembered between runs, and what it is remembered under
Keysevery key, the leader layers and the three styles
Spellingremarks checked against your own dictionaries
What comes backthe review as the agent receives it
Teaching the agentthe skill, and where each agent looks for it
Diffshow a diff is drawn, and the two engines behind it
Version controlthe git and jj profiles, and writing one for another VCS
Configurationthe four layers, the settings window, and trusting a repository's file

What it is for

Reviewing an agent's work in a terminal has two problems: the terminal, and the diff. The window is native — no webview, no browser engine, no multiplexer to be compatible with. And only the annotations travel back, so asking for a review on a large change costs a few hundred tokens instead of the patch.