Codiff 0.1 is out * Fast Local Code Reviews * Optional LLM Walkthroughs * Inline Review Comments This is the best companion for reviewing output of coding agents. macOS Release:

Great thing to work on! Been waiting for better code review tools

How can I make it even better?

I'd like the tool to highlight areas where my architectural review might actually be helpful skip the "yeah you might nitpick some syntax here but who cares" sections without me needing to read them to find out it's low value identify files that are very horizontal and impact multiple things vs. leafy stuff that has low blast radius basically anything that helps me do high leverage review and not block work where the review doesn't matter I mostly use GitHub's VSCode plugin now to review in the IDE so I can see full context, but it's buggy

I'm iterating on this today. Here is the initial new version: * Small summary * Tells you what matters * Deprioritizes boring stuff

that looks sweet

I'm pretty happy with this version now. Will ship it later today. I expect it'll need more iteration, but I think I'll use this 100 times a day.

"wide impact" is a great callout can I expand the diffs to look at the full file with the diff in it, or at least the surrounding area?

Not yet, but there is now an "Open" button that opens the file.

I’d make the review show two things separately: what the agent intended to change, and what actually changed in the diff. That catches a lot of the “looks fine until you read the side effects” failures.

Send a PR with your idea for it.

Fair. I’ll try it locally first. If the diff/intent split fits the current flow cleanly, I’ll send a small PR rather than armchair it here.

Fast local review is a strong companion to coding agents. The more agents write code, the more valuable it becomes to separate intent, actual diff, side effects, and reviewer comments in one tight loop.

@steipete Would you be interested by a PR that adds sending inline review comments to some agents sessions directly ? Maybe like a command to open the diff review from a session like plannotator does

The missing piece in agent code review is provenance. I want the review to say which agent wrote the diff, which tests ran, and which comments are blockers vs cleanup. Otherwise it becomes another noisy PR bot.

claude ships prs faster than i can read them. need this

日本語での詳細解説はこちら Check out my Japanese breakdown:

@lgrammel @coderabbitai please wake up

@lgrammel @coderabbitai Coderabbit is confused. It hurt itself in its confusion.

can you release it in arch linux

Sure:

Neat and pretty

Feel free to checkout orche as well which comes with not only reviewing tool but also a wrapper to create worktrees and dev environments programmatically, so that your sessions can be triggered by an orchestrating agent every time you start working on a new ticket:

Does it show full file tree of the project regardless of files changed in diff? Can I one click filter full tree to only changed/added files? Does it support data flow analysis and code navigation on semantic level so I can trace function call chains and see code quality metrics inline?

@steipete No it’s focused on diffs only right now. Looking at all files could be added. It doesn’t support the other stuff. Fork it and add it to your fork.

@steipete Nah, I don’t think it makes sense to replicate existing IDE features in lowq way. Have you thought about doing it other way around? I found it very interesting to understand why people IDE doesn’t fit for this?

@steipete IDE does too much. I don’t review code in the IDE. To be honest, I don’t use an IDE for much of anything anymore.

looks interesting. the window construction and UI are pretty rough, are you planning to make it more refined and Mac-looking?

more commits fewer tests. wild times

codiff + agent output is a real combo. the inline walkthrough part saves time you'd normally spend context-switching between terminal and code. llm walkthroughs especially.

the context switch cost is real — but usually the issue isn't the switching itself, it's that you're switching away from something to write a reply that has no actual situation to draw from. that's where generic happens

Finally a tool that actually makes reviewing agent code feel like less of a chore.

That's great news about Codiff 0.1, @cnakazawa. It sounds like a very useful tool for local code reviews. I'm curious, though, if it also works well for reviewing code written What are your thoughts on that?

i'd use this just to review agent output without opening a whole PR flow. local-first reviews feel underrated.

Not exactly sold on pre-commit vs after commit workflow. Getting functional via X commits and then review/refactor works better for me. Could be a git issue more than workflow issue tho

Next release can review the history too.

Fast Local Code Reviews is the useful bit. agent output needs review UI that stays inside the diff, not another chat transcript.

A local review layer is underrated. For agent-written code I want the first pass to answer boring questions: files touched, risky assumptions, tests actually run, and what still needs a human read.

this is awesome, thanks for releasing!

@steipete Will check this out

This is a very timely direction. The hard part with coding agents is no longer just getting a diff, it is making the handoff reviewable and easy to annotate. Local, fast, with comments you can copy back into the workflow feels exactly right. Excited to try it.

the reason most people aren't using ai to review their own code is that it's still hard to explain the why behind those suggestions. codiff 0.1 starts to bridge that gap.