CodeRabbit Commands: Every @coderabbitai Command (2026)
The complete CodeRabbit command reference - review, full review, pause, resume, resolve, ignore, summary, configuration, help - plus how to chat with the bot and when each one is the right call.
Published:
The command table
Every command is a pull request comment starting with @coderabbitai. This is the whole reference.
| Command | What it does | When you want it |
|---|---|---|
@coderabbitai review | Incremental review of what changed since last time | Automatic reviews are off, or you want a re-check after a push |
@coderabbitai full review | Re-reviews every file from scratch | After a rebase, force push, or large refactor |
@coderabbitai pause | Stops reviews on this PR | Mid-iteration, when you do not want a review per commit |
@coderabbitai resume | Restarts reviews | You are ready for feedback again |
@coderabbitai resolve | Marks all CodeRabbit comments resolved | You have addressed the feedback your own way |
@coderabbitai summary | Regenerates the PR summary | The description drifted after you changed scope |
@coderabbitai configuration | Prints the active config for the repo | Debugging why it is behaving unexpectedly |
@coderabbitai help | Lists commands | You forgot one |
Two that are not comments:
@coderabbitai ignoregoes in the pull request description. While that text is present, CodeRabbit will not auto-review any commit on the PR. You can still trigger a manual review with the commands above.@coderabbitaiin the pull request title makes CodeRabbit generate the title for you.
If your organisation runs the bot under a different service-account handle, substitute it. The keywords do not change.
The three commands that actually matter
Most of that table is situational. Three of them change how you work.
pause is the one people should use and do not
The default behaviour reviews every push. If you are iterating - push, fix, push, fix - you get a review per commit, which is where “CodeRabbit is too noisy” usually comes from. It is not the reviewer being verbose, it is you asking it eight times.
Comment @coderabbitai pause when you start iterating. Comment @coderabbitai resume when the branch is actually ready. You get one good review instead of eight partial ones, and the PR thread stays readable.
This is also the fix for the rate limit on the free tier, where each of those eight reviews counts against your hourly allowance. We covered that in is CodeRabbit free for open source.
full review after any history rewrite
Incremental review works from what changed since the last pass. Rebase, squash, or force push and that history no longer describes reality, so the incremental diff can be nonsense - reviewing changes that no longer exist or skipping ones that do.
Any time you rewrite history, @coderabbitai full review. It is the difference between a review of your branch and a review of a branch that used to exist.
resolve to close out a thread honestly
You will often fix a flagged issue differently from how it was suggested. Clicking resolve on fifteen individual comments is tedious enough that people just leave them open, and an open thread on a merged PR is noise for whoever reads it next.
@coderabbitai resolve clears them in one action.
Chatting with the bot
The commands are the deterministic surface. Underneath it, CodeRabbit is conversational, and this is the underused half of the product.
Reply to a review comment to continue that specific thread:
I pushed a fix in commit a1b2c3d, please review it.
Tag it inline on the Files changed tab to ask about a specific location:
@coderabbitai generate unit testing code for this file.
Tag it in a PR comment to ask about the branch as a whole:
@coderabbitai read src/utils.ts and generate unit testing code
@coderabbitai help me debug my CodeRabbit configuration file
That last one is worth knowing about. When reviews are behaving oddly, you can ask the bot to explain its own configuration rather than reading the YAML schema yourself.
The constraint is the context window. Large, vague requests degrade noticeably. “Refactor this service” gets you a worse answer than three scoped questions. Prefer inline comments for anything file-specific, since they carry the location as context for free.
The workflow that works
Putting it together, the pattern that produces the least friction:
- Open the PR as a draft if you are still iterating, or comment
@coderabbitai pauseimmediately. - Iterate freely. No reviews, no noise, no rate limit consumption.
- Comment
@coderabbitai resumewhen the branch is genuinely ready. - Read the review. Reply inline to anything you disagree with rather than silently ignoring it - the reply is context for the next pass.
- Push fixes, let the incremental review confirm them.
@coderabbitai resolveto clear anything you handled differently.@coderabbitai full reviewif you rebased at any point before merging.
Teams that fight with AI reviewers are usually running step 2 with reviews switched on.
Making it permanent with configuration
Commands control one pull request. For behaviour you want every time, use a .coderabbit.yaml file at the repository root.
Two keys pair naturally with the commands above:
ignored_branch- a comma-separated list of branches to skip, wildcards supported. Better than remembering@coderabbitai ignoreon every dependabot PR.ignored_titles- snippets that, when matched in a PR title, suppress review.WIPandchoreare the obvious entries.
Add the schema line at the top of the file and your editor’s YAML language server will autocomplete the rest:
# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
The full key reference is in our CodeRabbit configuration guide, and custom review instructions covers telling it what to care about in plain English.
Bottom line
The command surface is small and most of it you will use once. The three worth committing to memory are pause and resume around iteration, and full review after any rebase.
If CodeRabbit feels too chatty, try pausing during iteration before you go looking for a quieter tool. A large share of the volume people complain about is the reviewer being asked eight times rather than once. For the configuration-level fixes when that is not enough, see our CodeRabbit review for where it genuinely trades noise for coverage.
Further Reading
GitarComments are not enough
Gitar applies the fix, validates it in CI, and clears the queue.
See it on your repo Read our independent Gitar reviewFrequently Asked Questions
What are the CodeRabbit commands?
All commands start with @coderabbitai posted as a pull request comment. The core set is review, full review, pause, resume, resolve, summary, configuration, and help. Separately, @coderabbitai ignore goes in the PR description rather than a comment, and putting @coderabbitai in the PR title makes CodeRabbit generate the title for you.
How do I stop CodeRabbit reviewing a pull request?
Add @coderabbitai ignore anywhere in the pull request description. As long as that text stays in the description, CodeRabbit will not automatically review any commits on that PR. For a temporary stop instead, comment @coderabbitai pause and later @coderabbitai resume.
What is the difference between review and full review?
@coderabbitai review is incremental - it reviews what changed since the last review. @coderabbitai full review discards that history and re-reviews every file in the pull request from scratch. Use full review after a rebase, a force push, or a large refactor where incremental context has become misleading.
How do I make CodeRabbit resolve all its comments?
Comment @coderabbitai resolve on the pull request. It marks all outstanding CodeRabbit review comments as resolved in one action, which is the fastest way to clear a PR where you have addressed the feedback in a different way than suggested.
Can I ask CodeRabbit questions?
Yes. Tag @coderabbitai in a PR comment to ask about the branch, or tag it in an inline comment on the Files changed tab to ask about a specific location. It can read files, generate unit tests, produce diagrams, and explain its own configuration. Keep requests scoped - the context window is finite and large asks degrade badly.
Do the commands work if my bot has a different name?
Use whatever handle your installation uses. Self-hosted and enterprise installs sometimes run under a different service-account name, and the command keyword stays the same - only the mention changes.
Explore More
Tool Reviews
Related Articles
Free Newsletter
Stay ahead with AI dev tools
Weekly insights on AI code review, static analysis, and developer productivity. No spam, unsubscribe anytime.
Join developers getting weekly AI tool insights.
Related Articles
CodeRabbit Too Noisy? How to Cut the Comments (2026)
An independent audit found 36% of CodeRabbit's comments were nitpicks or useless. Here is the .coderabbit.yaml config, the workflow changes, and the review-frequency settings that fix it.
August 1, 2026
how-toHow to Give Code Review Feedback That Lands - Blocking vs Nitpick (2026)
Practical guidance on giving code review feedback - blocking comments vs nitpicks, tone that keeps authors receptive, and real before-and-after comment examples.
July 28, 2026
how-toHow to Review AI-Generated Code - A 2026 Checklist
A practical checklist for reviewing and gating AI-generated code from Copilot, Cursor, and Claude - what to scrutinize, why it slips past normal review, and tools that help.
July 28, 2026
CodeRabbit Review