Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.cubic.dev/llms.txt

Use this file to discover all available pages before exploring further.

The cubic CLI runs a lightweight version of cubic’s AI code review directly into your terminal. It’s designed as a fast pre-flight check before you push, not a replacement for cubic’s cloud PR review. The CLI detects bugs in your local changes and generates a prompt for you to paste into your AI IDE (like Cursor) to automatically fix them. It syncs with supported cubic cloud context such as your custom agents, past feedback, and repository settings, but it does not use the same models or review pipeline as cubic’s cloud PR review.
Do not expect identical results between cubic review and cubic’s cloud PR reviews. The CLI is intentionally faster and less thorough for local iteration, while cloud PR reviews are more thorough and may find issues that the CLI did not flag.
CLI usage shows up in your analytics dashboard as CLI sessions so you can track adoption across your team.

​
Getting started

Install the cubic CLI using one of the following methods: 
curl -fsSL https://cubic.dev/install | bash

​
npx

npx @cubic-dev-ai/cli install -g
After installation, type cubic to open the CLI and start a review. cubic CLI home screen showing review preset options

​
Authentication

The first time you run cubic, it will automatically open your browser to sign in with your cubic.dev account. Once authenticated, you’re all set — no extra steps needed. If you run into any authentication issues, press Ctrl+L to log out and re-authenticate.

​
Bring your coding subscription

The CLI includes a default model, but you can also connect your ChatGPT Plus/Pro or Claude Code account to run reviews with additional models. These integrations are optional.

​
Use your ChatGPT subscription

Connect your ChatGPT account to run reviews against Codex models with your existing plan: 
cubic auth connect codex
Sign-in happens in your browser. Once connected, choose the Codex model that fits the task.

​
Use your Claude Code account

Connect your Claude Code account to run reviews with Claude: 
cubic auth connect claude-code
Choose the Claude model that fits the review: opus when thoroughness matters most, sonnet for everyday reviews, haiku when you need speed.

​
Why this is permitted under Anthropic’s policy

cubic runs your local Claude Code as a subprocess and talks to it over the Agent Client Protocol (ACP), the same pattern Zed uses for its Claude Agent. Your Claude Code installation authenticates with Anthropic, makes every model call, and bills you directly. cubic never sees, stores, or proxies your credentials, and does not call Claude’s API, use the Agent SDK, or offer Claude.ai login. Reviews are interactive: you start each session and drive it from the UI. cubic does not run autonomous loops in the background. This is the “ordinary use of Claude Code” that Anthropic’s usage policy permits for Pro/Max subscribers.

​
Switch providers any time

Press Ctrl+P to switch provider or m to switch model. If you have more than one provider connected, cubic picks the best available model by default. You can change either at any time — no re-authentication needed. To disconnect a provider, run cubic auth logout and pick the one you want to remove. Disconnecting Claude Code only detaches it from cubic; your Claude Code login stays intact.

​
When to use local review

Use one of these two workflows:
  • Default: GitHub review only: Push your branch, let cubic review the PR on GitHub, then fix the comments. This is the simplest setup and the best default for most teams.
  • Advanced: local review plus GitHub review: Run local review before you push so you can iterate on your code before opening a PR, then rely on GitHub review as the final pass.

​
Use it with your coding agent

Most teams use the cubic CLI together with a coding agent:
  • Ask your coding agent to "review my code" while you are working.
  • Run cubic review when you want an explicit local pass in the terminal.
  • Re-run the local review after fixes, then push and rely on the GitHub PR review as the final pass.
If you have not connected cubic to your editor yet, start with Connect cubic to your IDE.

​
Add this to AGENTS.md, CLAUDE.md, or .cursorrules

Use this snippet if you want your coding agent to follow the default GitHub-first workflow: 
## cubic workflow

After pushing, wait for cubic's GitHub review.
Ask the agent to "check cubic comments" or "fix cubic issues".
Push the fixes and repeat until the pull request is clean.
Add this extra block if you want the advanced local-review loop: 
Before committing, run `cubic review` or ask the agent to "review my code".
Fix the validated issues and run the local review again.
Repeat until the local review is clean or only disputed issues remain.

Do not expect the cubic CLI and GitHub PR review to return identical results.
The CLI is faster and less thorough.

​
Common Workflows

The CLI adapts to different stages of your development process. cubic CLI review results showing prioritized issues

​
1. Review Uncommitted Changes (Default)

Review code you are currently working on, before you commit. 
cubic review

​
2. Review a Pull Request (Branch Review)

Compare your current branch against a base branch (e.g., main) to catch issues before opening a PR. 
# Auto-detect base branch
cubic review --base

# Explicit base branch
cubic review --base main

​
3. Review a Specific Commit

Analyze changes introduced by a specific commit hash or reference. 
cubic review --commit HEAD~1

​
4. Custom Instructions

Focus the review on specific concerns like security or performance. 
cubic review --prompt "check for XSS vulnerabilities"

​
Command Reference

The primary command is cubic review. 
cubic review [options]

​
Options

OptionAliasDescription
--json-jOutput results as JSON (for CI/automation)
--base [branch]-bReview against a base branch (PR-style). Auto-detects base if no branch specified
--commit <ref>-cReview a specific commit
--prompt <instructions>-pCustom review instructions
Note: --base, --commit, and --prompt are mutually exclusive.

​
Output Formats

​
Text Output (Default)

Human-readable output with colored priority labels, designed for interactive use. 
[P0] src/api/auth.ts:45
   SQL injection vulnerability in user lookup
   User input is concatenated directly into SQL query without parameterization.

​
JSON Output

Structured output for tools and automation. 
cubic review --json

{
  "issues": [
    {
      "priority": "P0",
      "file": "src/api/auth.ts",
      "line": 45,
      "title": "SQL injection vulnerability in user lookup",
      "description": "User input is concatenated directly into SQL query without parameterization."
    }
  ]
}

​
Auto-update

The cubic CLI updates itself automatically. Each time you run a command, it checks in the background for a newer version. If one is available, it upgrades the install in place and then runs that command on the updated binary.

​
Disabling auto-installs

Add to your global cubic config (~/.config/cubic/config.json): 
{
  "disable_auto_updates": true
}
With this set, cubic still tells you when a new version is available but does not install it: 
New version available: v1.2.2. Run `cubic upgrade` to install.

​
Skipping the check entirely

Set the environment variable to opt out of both the version check and any install: 
export CUBIC_DISABLE_AUTOUPDATE=1
This is useful in CI, on machines without outbound internet, or when you want to pin a specific version.

​
Upgrading manually

Run cubic upgrade to upgrade immediately to the latest release, or cubic upgrade <version> to install a specific version.

​
AI attribution tracking

The cubic CLI installs git-ai during setup to tag AI-assisted commits with git notes. These notes power the AI coding analytics dashboard for tool usage and authorship trends. Attribution tracking is enabled by default. git-ai shows a brief AI vs human authorship summary after each commit.

​
Opting out at install time

Set the CUBIC_DISABLE_GIT_AI environment variable to skip git-ai during installation: 
# curl
CUBIC_DISABLE_GIT_AI=true curl -fsSL https://cubic.dev/install | bash

# npx
CUBIC_DISABLE_GIT_AI=true npx @cubic-dev-ai/cli install -g

​
Managing tracking after installation

Use cubic stats to enable, disable, or check the status of AI attribution tracking at any time: 
# Check current status
cubic stats status

# Disable tracking
cubic stats disable

# Re-enable tracking
cubic stats enable
cubic stats disable removes git-ai hooks and persists your preference so future upgrades will not re-enable it. cubic stats enable re-installs git-ai and clears the opt-out flag.

​
Limitations

Usage is unlimited while the CLI is in alpha, subject to rate limits to ensure fair access for all users. Because the CLI is optimized for speed and local iteration, it is less thorough than cubic’s cloud PR reviews. If you need the most complete review, open or update a PR and rely on the cloud review as the source of truth.