slop-review
A browser-first code review tool for leaving inline diff comments and handing them to an LLM.

About
slop-review is a code review tool for commenting on git diffs and handing the threads to an LLM. It runs locally in your browser or in a terminal via Carbonyl, so you can review a branch, mark files, and keep the whole loop inside one machine. The app is built for review workflows rather than passive diff viewing, with inline threads stored as plain JSON under .reviews/ and no clipboard or HTTP API handoff required for the agent loop.
What is slop-review
slop-review is a local code review tool for working through a git branch, adding inline comments to the diff, and then passing those review threads to an LLM. It is designed around the review loop itself, not just around reading changes. You can use it on any git repo, and the same review UI runs either as a browser app or directly in a terminal through Carbonyl.
The app keeps review data as plain JSON in a .reviews/ folder. That means the review tool does not depend on a remote backend, a clipboard bridge, or a server-based agent API to move comments between you and your model. For teams or solo developers who want a lightweight review tool, that is the main appeal of slop-review.
Key features
slop-review focuses on a few specific pieces of the review tool workflow:
- Importance-ordered file list: changed files are sorted by how central they are to the change set, instead of alphabetically.
- Keyboard-driven review UI: the app is designed for single-key actions and Vim-like navigation.
- Inline comments on diffs: you can comment on the new side or the old side of a change, including single-line and multi-line ranges.
- LLM handoff loop: a bundled Claude Code skill can read and write
.reviews/JSON so an agent can act as reviewer or reviewee. - Blob-keyed reviewed marks: reviewed state is tied to a file blob SHA, so marks become invalid only when that file actually changes.
- Cross-file symbol lookup: double-click an identifier to see matching occurrences across changed files.
- Three diff modes: review a full branch, a single commit, or your current working copy.
- Two renderers: the same review tool works in a browser tab or in a terminal pane through Carbonyl.
The browser mode is the easiest starting point, while the terminal mode is useful when you want the review tool to live in one pane with your editor or shell.
How to use it
- Run the app in a git repository. The project documents a quick start with
npx slop-review. - Open the local UI on
localhost:9410in the browser, or launch it in terminal mode with--carbonyl. - Pick the diff mode you want: branch diff, commit diff, or working-copy diff.
- Move through files with the keyboard and let the importance ordering surface the most relevant changes first.
- Press
cto add an inline comment, or use range selection for multi-line feedback. - Mark files reviewed by clicking the file header when you are done with a file.
- If you want an agent to help, install the bundled skill and ask it to review or respond to the open threads.
The review tool exposes which-key style hints as you press keys, so only valid actions for the current state are shown.
Who it is for
slop-review is a good fit for developers who want a faster local review tool than a plain diff viewer. It is especially useful for people who:
- review their own branch before opening a PR
- want inline comments preserved as structured data
- use an LLM or coding agent to help address review notes
- prefer keyboard-heavy workflows over mouse-driven code review
- work in repos where a terminal-first setup is more convenient than a separate browser window
It is also a practical review tool for anyone experimenting with human-plus-agent review loops. Because the comments are stored locally, it can work as a private scratchpad for branch review before anything is shared elsewhere.
What to know before you use it
slop-review is intentionally narrow in scope, and that is worth understanding before you rely on it as a review tool.
- It is not a hosted code review platform like GitHub pull requests or GitLab merge requests.
- The LLM loop is based on local JSON files and a bundled Claude Code skill, not on a generic HTTP API.
- The app needs Node 20 or newer for the documented
npxworkflow. - Terminal mode depends on Carbonyl, so the terminal experience is not the same as a native TUI.
- Reviewed marks are keyed to blob SHA, which is helpful for accuracy, but a later rewrite of a file can clear a prior review mark.
- The tool can help you inspect diffs and manage threads, but it does not replace source control history, CI, or an official PR review in a hosted platform when that is required.
For branch-level review, local commenting, and LLM-assisted thread handling, slop-review is a focused review tool. When you need authoritative approvals, audit trails, or repository-hosted collaboration, use the official platform instead.
FAQ
Is slop-review free to use?
Yes, it is open source and documented with a zero-dependency local workflow. The source material does not list a paid tier or subscription.
How do I install and run slop-review?
Run `npx slop-review` in any git repository to start the local web server at localhost:9410. Alternatively, use Carbonyl to open it directly inside your terminal without a separate browser window.
Does slop-review run in the browser or terminal?
Both. The same review UI runs as a local web app in your browser and inside your terminal through Carbonyl.
Which LLMs does slop-review support?
It bundles a skill for Claude Code that can read and write the .reviews/ JSON directly. Because the data is plain JSON, you can adapt it for any coding agent or LLM that can handle file-based input/output.
How does the LLM workflow work?
The app uses plain JSON files in `.reviews/` and a bundled Claude Code skill so an agent can act as reviewer or reviewee without a server API.
Can I use slop-review offline or in terminal-only environments?
Yes, with Carbonyl you can run the full GUI inside your terminal, making it ideal for remote servers or systems without a graphical browser. The web interface works offline if you have the files locally.
What kind of git review does it support?
It supports full branch diffs, single commits, and working-copy diffs. You can review inline comments on the diff in each mode.
How does slop-review decide which files to show first?
Files are ranked by importance: first by incoming reference count (how many other changed files import them), then by modification status (modified, added, removed), and finally by file type (source before tests, docs, configs). This ensures you start with the code that has the widest impact.
Who is slop-review best for?
It is best for developers who want a keyboard-driven local code review tool and who may want to hand review threads to an LLM.