I am working on speeding up the review-refine loop of LLM generated artefacts between me and the LLM. i run my coding agent (claude code) in auto / full-permission mode, so it doesn’t stop for accept/reject — by the time it’s done i’ve got a pile of generated code, docs, sometimes ui to review. and the slow part for me is to refine it before push it to remote. i would copy-paste snippets to point at what i wanted changed, screenshot the ui, describe where the problem was in prose, one message at a time. its cumbersome.

so i built prereview to speed up that review-fix loop. you run it in your repo (or point it at a file, a dir, or a running dev server), click what’s wrong: a diff line, a markdown/html block, a region of an image, a box on a live site — and leave a comment. the comments go to a csv, your agent reads it and fixes things and tells you what it did via a comment or even posts suggestions which you can accept or reject. it ships with a claude code skill, but the handoff to llm is just an open csv protocol, so any llm cli can drive it. stuff that might interest people here: single static go binary, fully offline; it reviews docs/images/live ui as well as code; comments re-anchor when the file changes under them. it’s mit and still early but I use it everyday. Its here: https://github.com/livetemplate/prereview

I am using claude cli/tmux on a hetzner box and connecting to it via claude remote control. I have connected the box and my phone over telnet which allows me to view any UI work. Sometimes I do have to switch to my laptop for UI desktop layouts.

One gap which I kept running into on both mobile and desktop was refining the initial plan and then later refining the generated artifacts which involved lots of imprecise copy-paste. To scratch my own itch I built a review tool to improve the velocity of planning and refining generated artifacts. It has become my daily driver: https://github.com/livetemplate/prereview

:) We are all converging on the same problems and tooling to fix them.

Built this in the last couple days to review changes done by claude on a hetzner box(over tailscale on my phone) and leave comments before pushing to remote : https://github.com/livetemplate/prereview.

Runs everywhere and can has a claude skill: /prereview which starts a new instance for the current changes. Has a “Hand-off” button to signal claude to process the comments.

UI is responsive. I use it on my iphone with claude mobile app + /remote-control to leave comments before pushing.Still requires polish but core features work.

That’s awesome! I built this specifically as a GUI because it’s become too much code to effectively review in the terminal for me.

exactly. LLM code output is so high that it has become unreasonable to review the code in terminal or ide

Would love to see some screenshots in your readme of this!

Yeah the project is right out of the oven so will add some usage gifs shortly to the repo. Meanwhile here are some screenshots on imgur: https://imgur.com/a/zxZo651

that’s exactly the core value proposition of the tool I have been working on : https://github.com/livetemplate/tinkerdown

I am working on https://github.com/livetemplate/tinkerdown to leverage the the two effective outputs of LLMs : html and markdown . Still WIP

Good ideas. I am working on something similar but with a tiered complexity to fallback to html for complex use cases : https://github.com/livetemplate/tinkerdown, https://github.com/livetemplate/tinkerdown/blob/main/docs/gu...

excellent work. I have been using a vscode extension for this for an year.

Thanks! What extension were you using?

Recently started using: https://github.com/babonet/Markco (cant find the similar extension i used previously since it got uninstalled). I add inline comments in the markdown and I have a skill which uses the comments to update the plans and remove resolve comments.

The streamed execution idea is novel to me. Not sure what’s it significance ?

I have been working on something with a similar goal:

https://github.com/livetemplate/tinkerdown

The significance is responsiveness — instead of waiting for the LLM to finish generating the entire code block before anything happens, each statement executes as soon as it's complete. So API calls start, UIs render, and errors surface while the LLM is still streaming tokens.

Combined with a slot mechanism, complex UIs build up progressively — a skeleton appears first, then each section fills in as the LLM generates it.

I wrote a deeper dive on how the streaming execution works technically: https://fabian-kuebler.com/posts/streaming-ts-execution/

That is super cool. Sorry to be nitpicky but would really like to know your mental model: I didn’t understand from the blog why user waiting for a functional UI is a problem ? isn’t the partial streamed UI non-functional ?

I can see the value in early user verification and maybe interrupting the LLM to not proceed on an invalid path but I guess this is customer facing so not as valuable.

"In interactive assistants, that latency makes or breaks the experience." Why ? Because user might just jump off ?

(edited)

Maybe I am a bit overdramatic ;) For me this is mostly about user experience. If the agent creates a complex mini app, the user might have to wait 30 seconds. That's 30 seconds without feedback. It's way nicer to already see information appearing - especially if that information is helpful. Also the UI can be functional already, even if it's not 100% complete!

No that makes sense. Waiting for feedback might lead to churn. Pretty cool idea.

Add a video or a live demo, there's still too much friction on this readme.

Always Show then Ask.

and I meant to say: tinkerdown looks pretty cool!

Last time I wrote here was almost 6 years ago: https://honestmusings.wordpress.com. Not sure when I will write again.

The first step counter button is supposed to reload the page. The second step buttons update the page w/o reload

The second counter adds and subtracts values other than 1 and therefore seems broken.

That's because there are hundreds of people coming from hacker news clicking the button. if you just refresh the page you can see the number change. The current total seems to be saved server side