AI News HubLIVE
站內改寫3 分鐘閱讀

待翻譯:Show HN: GuidedReview, a Chrome extension that helps review AI generated PRs

AI 服務暫時不可用,以下為來源摘要,待恢復後補全翻譯:A better way for humans to review AI generated code Makes “reading code” wayyyy better: clusters PR diffs into review units, short summaries, keyboard-first. AI structures the walkthrough — you still decide. Install the…

來源Hacker News AI作者: nshntarora

AI 服務暫時不可用,以下為來源正文,待恢復後補全翻譯。

A better way for humans to review AI generated code Makes “reading code” wayyyy better: clusters PR diffs into review units, short summaries, keyboard-first. AI structures the walkthrough — you still decide. Install the extension⌘+IStar on GitHub⌘+G Free · Open source · Bring your own LLM key Product demo⌘+Vto play Why? Code review is hard, and it's only getting harder why.md AI agents are writing code for you, and you have PRs pending your review. When you installed the hot new code review agent, you imagined it would make the job easy. You'd click "approve" and move on to the next PR. I wish it were true. Unfortunately, code review agents are not a replacement for you (your job is safe, yay?). They're very useful for finding bugs and edge cases you've missed, but they lack something only you have — "taste." You have more context. You know the people, the business, and the product better than any coding agent. You know when an abstraction is unnecessary. You know what needs a comment. You know where to break the rules. And, for that... Nothing beats "reading" the code. Problem is, reading code is not easy. It's even harder if your tools are designed for storing code rather than reading it. AI has made writing code too easy. It's time we used it to make reading code easy too. We need an experience designed for humans — not to avoid AI, but to use it in just the right places (not too little, not too much). Reading the code matters more now because humans aren't writing it anymore. That's why I built Guided Review. — @nshntarora Features Built for humans who still have to read the code. clustered-changes.diff Review changes, not files GitHub gives you every changed file in alphabetical order and leaves you to work out how they connect. We use AI to cluster related changes into review units, so you read the change, not the file list. navigation.keys Read and navigate how you write Guided Review is keyboard-first. Browse units, comment, and submit without leaving the keys — same muscle memory as writing code. summaries.ai Summaries (because AI) Get a 2 line overview of every change. Sometimes it's really helpful, sometimes it's not. Take it with a grain of salt. Model companies apparently block your API keys if you do not implement summarization. Seriously. approved-tools.json Tools corporate security has already approved Bring the AI provider your company already approved — Claude, OpenAI, or Grok — and your own API key. Want to read the code that handles that key? It's open source (without VC funding), so you can read it, or fork it and point it somewhere else, or modify it for your own workflow. Please check with your employer if they've policies about this. If you're a 2 person startup, you're (probably) fine. Your code never touches our infrastructure The extension talks directly to your AI provider and GitHub. We never see your diffs, your keys, or your code — because there's nothing on our end to see them with. Star on GitHubBring your own LLM key FAQs The questions everyone asks before installing. faq.md So how does it actually work? Once the Chrome extension is installed, it adds a Start Guided Review button to the GitHub PR page. Click it and the review opens as an overlay on top of GitHub. We read the diff from the PR, send it to your configured AI provider to get review units and summaries, map those back to the diff, and show you the result. Which AI providers work? Claude (Anthropic), OpenAI, and Grok (xAI). You pick the provider and model in the extension's options pageand paste your own API key. We'll keep adding more providers — request one on the GitHub repo, or open a PR that adds yours. Will this run up my API bill? Starting a review makes one call to your provider to cluster the diff into review units and summarize them. Cost scales with how big the diff is and which model you picked — a small PR on a cheap model is fractions of a cent, a huge PR on a frontier model is not. You're paying your provider directly, so whatever it costs shows up on their dashboard, not ours. We never see it. So you don't track anything at all? Almost. This marketing site uses an analytics service for page views and a few CTA clicks. That's it. The extension doesn't track anything and doesn't talk to any third-party servers other than GitHub and your AI provider. Is it free forever? Forever is a long time. It is free until it can be maintained for free. If demand for the tool increases and it takes a lot of my bandwidth to maintain, I will introduce some way to be compensated for the extra time. But since it's open source, you can fork it if you find I'm not a good person. Who are you? I'm Nishant, here's my website, here's my Twitter/Xprofile, here's my LinkedIn, and here's my credit card number - 4242 4242 4242 4242 install.sh Get Guided Review Install the Chrome extension, add an LLM API key in options, and open any GitHub pull request. Click Start Guided Review to begin. Install the extension⌘+IStar on GitHub⌘+G