HN 표시: Crit – AI 생성 코드 및 계획 문서 검토를 위한 터미널 TUI

hackernews | | 🔬 연구
#ai #claude #review #tui #에이전트 #코드 리뷰
원문 출처: hackernews · Genesis Park에서 요약 및 분석

요약

Crit는 터미널 기반의 TUI 도구로, AI 생성 코드 변경 사항과 문서를 검토하기 위해 설계되었습니다. 이 도구는 특정 줄에 대한 인라인 주석을 남겨 AI 에이전트가 편집하도록 하며, 코드 검토 모드에서는 git 저장소의 변경된 파일을 비교하고 문서 검토 모드에서는 마크다운 파일을 지원합니다. CLI는 완전히 AI 에이전트와 연동되어 터미널 내에서 검토 주기 전체를 처리할 수 있습니다.

본문

TUI for reviewing AI-generated code and plans — built for human-in-the-loop agentic coding workflows. Read a plan or review code changes across multiple files, leave inline comments, and let Claude Code address your feedback automatically. Built for the human-in-the-loop workflow: Your agent the writes code or a plan, you review it in a TUI, your agent seamlessly reads your comments and makes changes. crit is available as a Claude Code plugin. Add the marketplace and install: /plugin marketplace add kevindutra/crit /plugin install crit Then use /crit:review in Claude Code. It will ask whether you want to review code changes or a document, open the TUI, and after you close it, Claude reads your comments and makes changes. go install github.com/kevindutra/crit/cmd/crit@latest Make sure $GOPATH/bin (defaults to ~/go/bin ) is in your PATH : export PATH="$PATH:$(go env GOPATH)/bin" If you prefer not to use the plugin, you can install the skill directly: crit setup-claude # Install globally (~/.claude/skills/) crit setup-claude --project # Install for current project only Then use /crit-review in Claude Code for document reviews, or /crit-code-review for multi-file code reviews. - Go 1.21+ for building from source - tmux — required for the Claude Code integration. Crit opens the review TUI in a tmux split pane next to Claude Code. If you're not already in tmux, start one before launching Claude Code: tmux new -s work # Now launch Claude Code inside this tmux session claude If you forget, crit will tell you — but the split-pane review won't work outside of tmux. crit review --code Detects changed files in your git repo and opens a tabbed TUI with syntax highlighting, diff markers, and inline commenting across all changed files. - Diffs against unstaged changes by default, falls back to HEAD~1 ormain - Green gutter markers highlight changed lines - Comments are aggregated across all files in the session # Get all code review comments as JSON crit status --code - Run crit review --code — crit detects changed files and opens the tabbed TUI - Navigate between files and leave inline comments on the changes - Quit the TUI — comments are saved to .crit/ crit status --code outputs all comments across files as JSON- Claude (or any tool) reads the comments and edits the files crit review docs/plans/my-plan.md Opens a full-screen terminal UI with syntax-highlighted markdown, a comment sidebar, and modal overlays for adding/editing comments. When running inside tmux, you can open the TUI in a side-by-side split pane: # Open review in a tmux split and return immediately crit review docs/plan.md --detach # Open review in a tmux split and block until it closes crit review docs/plan.md --detach --wait This is how the Claude Code skill invokes crit — --detach --wait is a single blocking call that opens the TUI next to Claude Code and waits for you to finish reviewing. - Claude writes a plan (or you open any markdown file) crit review opens the TUI — read through and leave inline comments- Comments are stored as JSON in a local .crit/ directory (gitignored by default) crit status outputs comments as JSON for Claude (or any tool) to consume- Claude reads the comments, edits the document, and you can re-review | Key | Action | |---|---| j / k | Scroll down / up | ctrl+d / ctrl+u | Half page down / up | g / G | Jump to top / bottom | enter | Add comment at current line | v | Visual select mode (multi-line comments) | s | Toggle comment sidebar | [ / ] | Jump to prev / next comment | q | Save & quit | Code review only: | Key | Action | |---|---| tab / shift+tab | Next / previous file tab | n / N | Jump to next / previous change | / | Search file tabs | # Add a comment programmatically crit comment docs/plan.md --line 15 --body "This needs more detail" # Multi-line comment crit comment docs/plan.md --line 10 --end-line 20 --body "Rethink this section" # Get review comments as JSON (single file) crit status docs/plan.md # Get all code review comments as JSON crit status --code # Bash crit completion bash > /etc/bash_completion.d/crit # Zsh crit completion zsh > "${fpath[1]}/_crit" # Fish crit completion fish > ~/.config/fish/completions/crit.fish go test ./... go build ./... go vet ./... MIT

Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.

공유

관련 저널 읽기

전체 보기 →