Reviewing a 1,000-line markdown plan that an AI agent just generated inside your terminal is a miserable experience. You scroll, lose your place, copy a chunk of text, switch back to the chat, and hope the agent figures out which part you meant. If it rewrites the file, you start the whole process over.
Crit is a small open-source CLI that tries to fix this. Run it on any file and it opens in your browser with a GitHub-style diff view. You click on specific lines, leave inline comments, and hit Finish. Crit then builds a structured prompt from your annotations and copies it to your clipboard. Paste it back into your AI agent's chat, and the agent gets precise, line-referenced feedback instead of vague instructions.
The tool ships as a single binary with no dependencies. It works with any AI coding agent since the output is just a clipboard prompt, not a proprietary integration. The workflow is intentionally simple: file in, annotated feedback out.
This solves a real friction point. Anyone who has used Claude Code, Cursor, or similar agent-based coding tools knows the pain of trying to give specific feedback on large generated files. Saying "change the third bullet under the deployment section" in a chat window is ambiguous. Clicking directly on line 47 and typing "use Docker instead" is not.
Crit is free and available on GitHub. It is early-stage software, so expect rough edges, but the core idea is sound. As AI agents produce longer and more complex outputs, the review step becomes the bottleneck. Tools that make that review loop faster and more precise will matter more than most people realize right now.