tomasz-tomczyk/crit

What it solves

Crit provides a specialized review interface for interacting with AI agents. Instead of reviewing generated plans or code changes in a chat window, Crit allows users to point at specific lines of text, markdown, or live web pages to leave feedback that agents can then act upon.

How it works

Crit runs as a local binary that launches a web-based review UI. It supports several modes of review:

• Plan Review: Renders markdown files with a review interface.
• Code Review: Auto-detects git changes and shows syntax-highlighted diffs.
• Code/UI Review: Proxies a running development server or static HTML file to allow inline commenting on the frontend.
• Story Mode: Creates a thematic, chaptered overview of large diffs to help users understand the shape of changes before a line-by-line review.

Feedback is stored locally in review files. Users can sync these comments bidirectionally with GitHub PRs and GitLab MRs, or use the "Send to agent" feature to pipe feedback directly to a configured AI agent (like Claude Code, Aider, or Cline) for real-time fixes.

Who it’s for

Developers who use AI coding agents and want a more precise, visual way to provide feedback and approve changes rather than relying on chat-based instructions.

Highlights

• Adaptive UI: Different interfaces for markdown plans, git diffs, and live web applications.
• Agent Integration: Direct piping of comments to agents via agent_cmd for real-time iterations.
• Crosspoint Sync: Bidirectional synchronization with GitHub Pull Requests and GitLab Merge Requests.
• Crosspoint Sync: Bidirectional synchronization with GitHub Pull Requests and GitLab Merge Requests.
• Async Review: Ability to share reviews via public URLs for second opinions.
• Local-first: All review data is stored locally by default, with no telemetry or tracking.