Feature

Inline Comments.

Click and drag line numbers to select a range, then leave a comment - just like a pull request review, but on any file your agent touched.

1 # Implement auth system

2

3 ## Backend

4 Use JWT tokens for sessions.

alice · Line 4 Edit · Delete

Consider OAuth2 instead - simpler for users, no token refresh logic needed.

bob · Line 4 Edit · Delete

+1 on OAuth2. See the


                passport.js

docs for examples.

5 Rate limit all endpoints.

6 Store passwords with bcrypt.

alice · Line 6 Edit · Delete

bcrypt is fine but


                argon2id

is the current recommendation.

Multi-author color coding is available on shared reviews at crit.md.

How it works

1

Select a single line or drag across multiple lines to highlight a range. A comment form appears immediately so you can start typing.

2

Comments support markdown - use **bold**, *italic*, inline code, code blocks, links, and lists to format your feedback.

3

Edit or delete any comment you've made. Each comment is anchored to the exact lines it references.

4

On the hosted version (crit.md), multiple reviewers can comment on the same document. Each author's comments are color-coded by a unique hue so you can tell at a glance who said what.

Why this matters

AI coding agents are fast, but they're also opaque. When Claude Code or Cursor rewrites a function, you don't get a natural place to say "this is wrong at line 12" unless you're already in a PR workflow. Most people end up pasting chunks of code into the chat and hoping the agent understands which part they mean. That's slow and error-prone.

Inline comments let you anchor feedback to the exact lines that need it. Instead of writing "the error handling in the auth function looks wrong," you select lines 34-41 and leave a comment there. The agent reads .crit.json (the review state file) and knows precisely what you're reacting to. No ambiguity, no re-explaining context.

This also matters when reviewing plans, not just code. If your agent writes a markdown spec or a step-by-step plan before executing, you can review that document the same way you'd review a PR. Comment on the parts that need changing, approve the parts that look right, and hand it back.

How Crit compares

GitHub PR reviews have the same interaction model, but they require a repo, a branch, and a push — which doesn't fit a local planning loop where the document hasn't been committed yet. CodeRabbit and similar tools automate the review itself; Crit is for when you are the reviewer and you want structured control over what the agent does next.

Try it out

Install Crit and start reviewing agent output in seconds.

See live demo Star on GitHub

$ brew install tomasz-tomczyk/tap/crit

then run

$ crit or crit plan.md

Or download a pre-built binary from GitHub Releases.

More features

Git Diffs & Round Diffs AI Review Loop Vim Keybindings Share Reviews Syntax Highlighting Mermaid Diagrams