Hunk
GitHub
Hunk is a new terminal app for reviewing code diffs, especially ones written by AI coding agents. It turns raw git diffs into an interactive, navigable review UI right in your terminal.
Based on reporting by GitHub — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Code review tools built for terminals usually stop at syntax highlighting and calling it a day. Hunk, an open-source project from Modem, wants more than that. It's a review-first diff viewer that treats a changeset like something you actually explore, not just scroll past, and it leans hard into a world where AI agents are writing a growing share of the code that humans need to check.
The core idea is a multi-file review stream with sidebar navigation, plus split, stack, and responsive auto layouts that adapt to your terminal size. You can point it at a live git repo with hunk diff, review a specific commit with hunk show, or even feed it two raw files or a patch piped in from stdin. Watch mode keeps things live, auto-reloading as your working tree changes, using filesystem events where possible and falling back to polling when it can't. Jujutsu and Sapling users get native revset support too, since Hunk auto-detects those VCS setups instead of forcing everyone into git-shaped workflows.
What sets Hunk apart from tools like difftastic, delta, or diff-so-fancy is the agent-native layer bolted on top. It supports inline AI and agent annotations sitting right beside the code, and there's a skill file you can hand to an AI agent so it can annotate a live Hunk session directly. That's a fairly explicit bet: as more commits get generated by Claude, Copilot, or whatever agent a team has wired into its workflow, the bottleneck shifts from writing code to reviewing it, and plain-text diffs in a pager start to feel inadequate for that job.
Beyond the review UI, Hunk ships a genuinely deep configuration system. Themes can inherit from built-ins and override individual syntax scopes using Shiki/TextMate selectors, there's an experimental TypeScript extension API for rewriting changesets or adding VCS backends, and it can slot in as your git, jj, or sl pager with one config line. None of that is flashy, but it signals a project aiming to be a daily driver rather than a demo.
Hunk requires Node 18+, runs on macOS, Linux, and Windows, and is built on OpenTUI, which it also exposes as an embeddable component for other terminal apps. It's MIT licensed and sponsored by Modem, and the feature comparison table it publishes against rivals is blunt enough that it's clearly picking a fight over the review-UI niche specifically, not trying to out-diff difftastic on structural comparison.
My take — AI-written commentary, not fact-checked reporting
I like tools that admit the obvious: a huge chunk of code review is now reviewing AI output, and pretending your diff tool doesn't need to know that is silly. Hunk baking agent annotations and a skill file into the core workflow, rather than bolting AI on as a chatbot sidebar, is the right instinct. Whether it beats lumen or just fragments an already crowded terminal-diff space is the real question, but I'd rather see five scrappy open-source attempts at this than one closed SaaS dashboard nobody can pipe a patch into.
Read more about this at: GitHub
