Skip to content
Critique/docs
Platform

@critique Bot

Talk to Critique on GitHub pull requests — questions, explanations, and on-demand review commands.

Once the GitHub App is installed, you can @mention Critique in PR conversation threads. The bot reads the PR diff, description, thread history, and (for inline comments) the file and line context before replying.

Mention syntax

Use @critique (or your app’s slug if you use a custom GitHub App name). Do not include [bot] in the mention — GitHub adds that badge automatically.

@critique why did we switch from useEffect to useSyncExternalStore here?

Works in PR comments and inline review comments on specific lines.

Commands

SyntaxWhat it does
@critique <question>Ask anything about the PR or diff
@critique /reviewRun the full review pipeline (uses credits)
@critique /ask <question>Same as a plain mention
@critique /explainExplain the PR or the code at the commented line
@critique /fixSuggest a concrete fix for the discussed issue
@critique /improveSuggest quality or architecture improvements
@critique /securitySecurity-focused pass on the diff
@critique /testsSuggest missing or stronger tests
@critique /helpShow command help in the thread

/review is the only command that reruns the full automated pipeline (Scout → specialists → synthesis). Other commands use a single focused model call.

What the bot knows

For each mention, Critique assembles fresh context:

  • PR diff (within configured file limits)
  • Title, description, and comment thread
  • Selected repository context when indexing is available
  • For inline comments: path, hunk, and line

Mentions are independent — there is no hidden multi-turn session between separate comments on the same PR.

Configuration

Default mention names work for the hosted Critique app. Self-hosted or white-label installations may use a different bot slug; your admin will tell you which @name to use.

Chat model choice for PR comments is managed by Critique operators — you do not need to configure models in GitHub.

Rate and cost

There is no small fixed cap on mentions per PR, but each mention is a real model call. Very long threads increase context size and latency. Prefer /review only when you need a full gate rerun.

See Billing & credits.