coderabbit with every installation method; the install script and Windows installer also provide the cr shortcut. Run coderabbit <command> --help to see the same information in your terminal.
Global usage
CodeRabbit CLI for code review and cloud coding. Usage:coderabbit [options] [command]
Commands
auth
Authentication commands Usage:coderabbit auth [options] [command]
Examples
auth login
Authenticate with CodeRabbit Usage:coderabbit auth login [options]
Examples
--agent is only supported for localhost browser callbacks and never accepts pasted OAuth tokens.auth logout
Clear local authentication Usage:coderabbit auth logout [options]
Examples
auth status
Show current authentication status Usage:coderabbit auth status [options]
Examples
auth org
Switch the active organization Usage:coderabbit auth org [options]
Examples
--id requires a saved organization ID. Use —id=<value> for an ID beginning with a dash. Selection uses only the inventory saved by your existing login; it does not discover access or switch accounts, regions or instances. Normal sign-in checks may refresh saved tokens. An unknown or uncached ID fails. Without --id, --agent remains a read-only names-only list.
review
AI-driven code review for the current Git workspace Usage:coderabbit review [options] [command]
A review that fails or doesn’t cover every file exits with code 1, even after printing findings; see Exit codes. For which changes each scope flag selects, see Local review.
Examples
- Plain text is the default review mode.
- Use
coderabbit auth login --agentfor agent-driven OAuth login. --remoterequires--baseand--source-branch, and runs outside any git checkout.
review findings
Show the most recent local review with stored findings Usage:coderabbit review findings [options]
Examples
- A newer review with no stored findings does not hide older findings.
- Clearing is scoped to the review directory, current branch, and base branch.
- Findings stored under any other scope are left unchanged.
pullrequest
Read existing CodeRabbit output without starting a review. Requires exactly one of--show-prompts or --show-threads.
Usage: coderabbit pullrequest [options] <number-or-url>
This command reads CodeRabbit’s existing output for the pull request and does not start a review. Pass exactly one of --show-prompts or --show-threads: the command fails without one, and the two cannot be combined. If no prompt is found, check that reviews.enable_prompt_for_ai_agents is enabled in the configuration and that CodeRabbit has finished reviewing the pull request.
The command needs a CodeRabbit SaaS login or a stored API key, and self-hosted logins are not supported. Unlike coderabbit review, it does not start sign-in for you: if you have neither, run coderabbit auth login first.
Examples
- CodeRabbit SaaS sign-in or a stored agentic API key is required; the repository must be installed in the active organization.
- Only github.com pull requests are supported. A full URL works outside a git checkout.
stats
Show statistics from locally stored review history Usage:coderabbit stats [options]
Examples
api
Call the CodeRabbit Platform API Usage:coderabbit api [options] [command]
api request
Call any CodeRabbit API endpoint and print its JSON response Usage:coderabbit api request [options] <endpoint>
api me
Print the API key owner and selected organization as JSON Usage:coderabbit api me [options]
api repositories
Read accessible installed repositories as JSON Usage:coderabbit api repositories [options] [command]
api repositories list
Usage:coderabbit api repositories list [options]
api repositories get
Usage:coderabbit api repositories get [options] <repository-id>
api pull-requests
Read repository pull requests and current review observations as JSON Usage:coderabbit api pull-requests [options] [command]
api pull-requests list
Usage:coderabbit api pull-requests list [options] <repository-id>
api pull-requests get
Usage:coderabbit api pull-requests get [options] <repository-id> <number>
api pull-requests review-status
Read the existing CodeRabbit review observation as JSON Usage:coderabbit api pull-requests review-status [options] <repository-id> <number>
usage
Show included-review availability and billing-period usage Usage:coderabbit usage [options]
Examples
update
Update CodeRabbit CLI to the latest version Usage:coderabbit update [options]
When a newer version is available and Homebrew manages your installation, coderabbit update installs nothing. It prints the new version and the brew upgrade coderabbit command, then exits with status 0, so a script can’t treat a successful exit as proof that the CLI was updated. On Windows, the update finishes after the command exits; see Update CodeRabbit CLI.
Examples
Homebrew installations must be updated with
brew upgrade coderabbit.config
Create or configure repository settings through a guided flow Usage:coderabbit config [options] [command]
Examples
coderabbit:coderabbit-config).
Schema: Eligible hosted repositories: https://www.coderabbit.ai/integrations/schema.v3.json Other new configurations use V2, as do eligible repositories when the V3 schema is unavailable. Existing V2 configurations: https://www.coderabbit.ai/integrations/schema.v2.json
doctor
Check CodeRabbit CLI installation and local review readiness Usage:coderabbit doctor [options]
skills
Install, update, or uninstall CodeRabbit skills for detected coding agents. Usage:coderabbit skills [options] [command]
Examples
--agent on a subcommand to preview a read-only plan for approval, then run its exact action_required command.
skills install
Install missing skills and update existing skills Usage:coderabbit skills install [options]
Agent flow: run
coderabbit skills install --agent, obtain approval for the complete plan, then run the exact command in the action_required record.
skills update
Update existing skills without installing missing skills Usage:coderabbit skills update [options]
Agent flow: run
coderabbit skills update --agent, obtain approval for the complete plan, then run the exact command in the action_required record.
skills uninstall
Uninstall skills owned by the CodeRabbit CLI Usage:coderabbit skills uninstall [options]
Marketplace plugins, npx-managed skills, project copies, and other externally owned skills are never removed.
code
Use the same Coding Agent and skill library as the web app. Coding Agent commands are in beta. Usage:coderabbit code [options] [command]
Every coderabbit code command needs a CodeRabbit SaaS user login and acts in your selected organization. API key, self-hosted, and SSO workspace logins are not supported (requirements). If no organization is selected, run coderabbit auth org in an interactive terminal first. To learn what the cloud Coding Agent does with the work you send it, see Coding Agent.
code handoff
Create a new CodeRabbit Cloud task with a summary, optional plan, and automatically discovered session transcript Usage:coderabbit code handoff [options]
--summary is required. Run the command from a clean checkout of a branch that has an origin remote. Untracked files count as changes, so keep the summary and plan files outside the repository or commit them first. For what is uploaded and what happens next, see Continue a local session in the cloud.
code ls
List one page of cloud tasks, most recently updated first, in the selected organization. Use--cursor with the returned next cursor and the same filters to continue. Only your tasks are listed unless you pass --all; --all does not fetch all pages. Archived tasks are excluded unless you pass --include-archived. This is a live listing, not a snapshot.
Usage: coderabbit code ls [options]
code show
Show the status, repository, delivery, and pull request health of a cloud task. Usage:coderabbit code show [options] <task>
code permissions
Read saved and active-turn access. Changes apply to subsequent turns through the existing revocation policy; this command never restarts or resends work. Full access requires your verified GitHub grant and does not bypass branch protection. Usage:coderabbit code permissions [options] <task>
code settings
Read saved settings by default. Pass exactly one setting to change it without starting or restarting work. Settings can change while a task runs. Usage:coderabbit code settings [options] <task>
code pin
Save a task in your personal pinned list. Find your pinned tasks with code ls--pinned; add --all to include tasks owned by other people.
Usage: coderabbit code pin [options] <task>
code unpin
Remove a task from your personal pinned list. Usage:coderabbit code unpin [options] <task>
code notifications
Select your organization first withcoderabbit auth org. Use a task from that active organization; a task from another organization fails. These preferences apply to your Coding Agent tasks across the active organization, not just this task. Without --all-events, show whether any event type is enabled; this cannot identify individual enabled types. --all-events on or off sets all five event types: turn end, input needed, pull request created, commit pushed and plan ready. Enabling requires your existing Slack access and linked account. No task turn is started.
Usage: coderabbit code notifications [options] <task>
code new
Start a cloud task on the current branch of the Git repository in this directory. The task starts from that branch on origin. Local commits that are not pushed and uncommitted changes are not included. Usage:coderabbit code new [options] [prompt]
code goal
Read the goal by default. Setting an objective preserves paused/limited states; editing a completed goal restarts it. Clearing removes the goal. Use code cancel to stop a turn and code resume-m to send instructions.
Usage: coderabbit code goal [options] <task>
code resume
Open a cloud task in the terminal. Shows its trace and live updates, and lets you send messages, steer the running turn, answer agent questions, and stop the turn. Closing the view does not stop the task. With--agent, send at most one message, steer, or answer, then stream newline-delimited JSON until the turn ends, the agent asks a blocking question, or 9 minutes pass. Add --interactive for a persistent bidirectional NDJSON session: send message, steer, answer, cancel, show, or close requests on stdin; receive correlated responses and live trace events on stdout. EOF detaches without stopping the task.
Usage: coderabbit code resume [options] [task]
code queue
List queued messages and the current pause ID. To resume work that a Stop paused, pass--resume with --expected-pause-id from this list. The pause is checked again before resuming. Resuming uses the existing queued messages and billing; this command does not wait for their turns to finish.
Usage: coderabbit code queue [options] <task>
code cancel
Stop the running turn. Queued messages do not run after it: the server pauses or drops them. The task stays open: a new message starts a new turn. Usage:coderabbit code cancel [options] <task>
code archive
Archive one task without prompting: hide it from the active task list, make it read-only, and pause its schedules. It can be unarchived later. Running, starting, or stopping turns must finish first. Repeating an archive succeeds. This does not cancel turns or delete tasks. Usage:coderabbit code archive [options] <task>
code unarchive
Unarchive one task without prompting. This does not start a turn or resume paused schedules. Workspace and Automation tasks are not supported. Usage:coderabbit code unarchive [options] <task>
code ask
Ask the task’s side chat a question and print the answer. A side chat reads the task’s conversation and repository. It does not change the task, and the turn of the task keeps running. Each question is billed. The side chat stays open for follow-up questions: pass--follow-up with the turn ID of an earlier answer to keep its context. A task can have 3 open side chats. An idle side chat expires after 24 hours. Use code side-chats to list or close existing side chats. The command waits up to 3 minutes for the task’s environment and up to 15 minutes for the answer.
Usage: coderabbit code ask [options] <task> <question>
code outputs
List generated output files. Pass--save with a server path to save that file’s basename in the current directory. Existing files are never overwritten. Agent output contains metadata, not file contents.
Usage: coderabbit code outputs [options] <task>
code side-chats
List existing side chats without asking a question. Pass--close with a listed side-chat ID to close it. Closing can continue after this command returns; the output reports the server’s current status. The task’s turn keeps running.
Usage: coderabbit code side-chats [options] <task>
code schedules
List a task’s schedules. To delete one, pass its ID, observed revision and--yes. Deleting stops later runs; already dispatched work can still run. This command cannot create or edit schedules.
Usage: coderabbit code schedules [options] <task>
code plan
Show the latest finalized plan of a cloud task. With--approve, record your approval of that plan version. You cannot approve the plan of a task that you started. With --implement, request implementation of the selected version. Approval alone does not start implementation.
Usage: coderabbit code plan [options] <task>
code fix
Request a new billed turn to fix CI, address reviewcomments, or resolve merge conflicts on the task’s existing PR. This command does not wait for the repair to finish. Existing task permissions and billing apply.
Usage: coderabbit code fix [options] <task> <kind>
code push
Deliver the ready changes of a cloud task and wait until the delivery finishes. By default, CodeRabbit commits and pushes the changes, as the web app’s Commit and push action does. GitHub can first ask you to authorize CodeRabbit in a browser that is signed in to CodeRabbit. Usage:coderabbit code push [options] <task>
code autofix
Autofix publishes the task’s changes when the agent finishes, then fixes CodeRabbit findings, required CI, and merge conflicts.on starts it, off stops it, resume starts a new repair episode after a pause, and status shows its state. Activation does not confirm a repair turn started; a failed task may need to be retried or continued before resuming Autofix. recover resolves a pending delivery conflict while preserving unpublished work; resume after recovery finishes. Pass exactly one target: a task ID or URL, or --pr <number|url>. on, resume and recover check Coding Agent billing; on and resume with --pr can create an Autofix task for the pull request.
Usage: coderabbit code autofix|autopilot [options] <action> [task]
code skills
Manage cloud Coding Agent skills Usage:coderabbit code skills [options] [command]
code skills list
List active skills available to you in the selected organization. Use the skill ID and version ID to run an exact version. Usage:coderabbit code skills list [options]
code skills run
Request a new billed turn with the selected library skill ID and immutable version ID. It starts when the task queue allows it; this command does not wait for completion or resume a paused queue. No skill name or latest version is inferred. Usage:coderabbit code skills run [options] <task> <skill-id> <version-id> [message]
code skills import
Choose an agent, select local skills, and review their import into the cloud library. Pass a skill folder or SKILL.md to import one directly. With--yes and no --scope, new skills are personal within the selected organization and existing skills keep their visibility. Importing an owned skill with the same name publishes a new version. Noninteractive imports require a path and --yes.
Usage: coderabbit code skills import [options] [path]
completion
Print the shell completion script forcoderabbit and cr. It completes commands, options, and the task IDs that this computer listed or opened.
Usage: coderabbit completion [options] <shell>
Examples
- The script completes
coderabbitand thecralias. - Task IDs come from the tasks that this computer listed or opened with
coderabbit code. A Tab press makes no network call.