Kodus CLI

Catch bugs before they reach your pull request — AI code review from the terminal.

npm install -g @kodus/cli

Quick Start

# 1. Install
npm install -g @kodus/cli

# 2. Authenticate (or skip for trial mode — no account needed)
kodus auth login

# 3. Review your code
kodus review

That's it. Kodus analyzes your changes, finds issues, and lets you fix them interactively — or auto-fix everything at once with kodus review --fix.

What It Does

Code Review

Analyze local changes, staged files, commits, or branch diffs. Kodus finds bugs, security issues, performance problems, and style violations — then suggests fixes with real code.

kodus review                    # Review working tree changes (interactive)
kodus review --staged           # Only staged files
kodus review --branch main      # Compare against a branch
kodus review --fix              # Auto-apply all fixable issues
kodus review --prompt-only      # Structured output for AI agents

Reviews are context-aware — Kodus reads your .cursorrules, claude.md, and .kodus.md so suggestions follow your team's standards. More on review modes

PR Suggestions

Fetch AI-powered suggestions for open pull requests directly from your terminal.

kodus pr suggestions --pr-url https://github.com/org/repo/pull/42
kodus pr suggestions --pr-number 42 --repo-id <id>

Filter by severity, export as JSON or Markdown, or pipe into an AI agent with --prompt-only for automated fixes.

Decision Memory

AI agents make dozens of decisions per session — architecture choices, trade-offs, why approach X was picked over Y. Without a record, that reasoning vanishes when the session ends.

Kodus captures agent decisions into your repo as structured markdown. When you or another agent return to the code, the full context is there.

kodus decisions enable           # Install hooks + initialize config
kodus decisions status           # See what's been captured
kodus decisions show [name]      # View PR or module memory
kodus decisions promote          # Promote decisions to long-term memory

Stored in .kody/pr/by-sha/<head-sha>.md — versioned with your code, readable by humans and agents. More on decision memory

Best With AI Agents

Kodus is designed to work inside AI coding agents. While you can use it standalone, the real power comes when your agent runs reviews automatically and fixes issues in a loop — no manual intervention needed.

Works with: Claude Code, Cursor, Windsurf, GitHub Copilot, Gemini CLI, and 20+ more environments.

Install the Skill (recommended)

The fastest way to get started. Auto-detects your installed IDEs and sets everything up:

curl -fsSL https://review-skill.com/install | bash

This installs the Kodus CLI globally and deploys the review skill into every supported agent on your machine — Claude Code, Cursor, Windsurf, and others. One command, all environments.

How It Works With Agents

Once installed, your AI agent can autonomously:

Write code as usual
Run kodus review --prompt-only to analyze changes
Read the structured output and understand each issue
Fix the issues automatically
Repeat until the review is clean

This creates a tight feedback loop: the agent writes, reviews, and fixes — all without leaving your IDE.

Beyond reviews, Kodus also captures what your agent decided and why via Decision Memory. Every reasoning step is saved into your repo — so when you (or another agent) pick up the work later, the full context is already there. No more re-explaining what was done or losing decisions between sessions.

Setup: Claude Code

Add to your project's CLAUDE.md:

## Code Review
After implementing changes, run `kodus review --prompt-only` to check for issues.
If issues are found, fix them and re-run until clean.

Or use the skill directly — after installing via the command above, just ask Claude Code to review your code and it will use Kodus automatically.

Setup: Cursor / Windsurf

Add to your .cursorrules or equivalent:

When writing code:
1. Implement the feature
2. Run: kodus review --prompt-only
3. If issues are found, fix them automatically
4. Repeat until review is clean
5. Show final result

Setup: Headless / Shared Environments

Set a team key so agents and shared machines are authenticated without individual logins:

export KODUS_TEAM_KEY=kodus_xxxxx
kodus review --prompt-only

Works with Codex, CI runners, remote dev environments, and any context where personal login isn't practical. Get your key at app.kodus.io/settings/cli.

Copy & Paste Workflow (interactive)

If you prefer manual control:

Run kodus review
Navigate to a file with issues
Select "Copy fix prompt for AI agent"
Paste into Claude Code or Cursor — the AI fixes it

The copied prompt includes file path, line numbers, severity, and detailed suggestions — optimized for AI agents.

Installation

Skill installer (recommended — CLI + all your agents)

curl -fsSL https://review-skill.com/install | bash

Installs the CLI and deploys the review skill to all detected agents in one step.

CLI only

npm

npm install -g @kodus/cli

npx (no install)

npx @kodus/cli review

curl

curl -fsSL https://raw.githubusercontent.com/kodustech/cli/main/install.sh | bash

Homebrew (coming soon)

brew install kodus/tap/kodus

Review Modes

Interactive (default)

kodus review

Navigate files with issue counts, preview fixes before applying, and copy AI-friendly prompts to paste into Claude Code or Cursor.

Auto-fix

kodus review --fix

Applies all fixable issues at once. Shows a confirmation prompt before making changes.

AI Agent

kodus review --prompt-only

Minimal, structured output designed for Claude Code, Cursor, and Windsurf. Perfect for autonomous generate-review-fix loops.

More: output formats & flags

Output Formats

kodus review                           # Interactive (default)
kodus review --format json             # JSON output
kodus review --format markdown         # Markdown report
kodus review --prompt-only             # AI agent output
kodus review --format markdown -o report.md  # Save to file

Diff Targets

kodus review                           # Working tree changes
kodus review --staged                  # Staged files only
kodus review --commit HEAD~1           # Specific commit
kodus review --branch main             # Compare against branch
kodus review src/index.ts src/utils.ts # Specific files

All Flags

Flag	Description
`--staged`	Analyze only staged files
`--commit <sha>`	Analyze a specific commit
`--branch <name>`	Compare against a branch
`--rules-only`	Only check configured rules
`--fast`	Faster analysis for large diffs
`--fix`	Auto-apply all fixable issues
`--prompt-only`	AI agent optimized output
`--context <file>`	Include custom context file
`--format <fmt>`	Output format: `terminal`, `json`, `markdown`
`--output <file>`	Save output to file
`--fail-on <severity>`	Exit code 1 if issues meet or exceed severity
`-i, --interactive`	Explicitly enable interactive mode

Decision Memory

Full reference for the decision capture system (intro above).

# Enable with specific agents
kodus decisions enable --agents claude,cursor,codex

# Custom Codex config path
kodus decisions enable --agents codex --codex-config ~/.codex/config.toml

# Overwrite existing config
kodus decisions enable --force

# Check what's been captured on current branch
kodus decisions status

# View decisions for a PR or specific module
kodus decisions show [name]

# Promote PR-level decisions to long-term module memory
kodus decisions promote --branch feat/auth --modules auth,users

# Disable hooks (preserves all captured data in .kody/)
kodus decisions disable

How it works: Hooks fire on agent turn-complete events and persist decisions to .kody/pr/by-sha/<head-sha>.md. Files are committed to your repo, versioned with your code, readable by humans and agents.

Supported agents: Claude Code, Cursor, Codex.

CI/CD & Git Hooks

Pre-push Hook

kodus hook install --fail-on error   # Block pushes with errors
kodus hook status                     # Check hook status
kodus hook uninstall                  # Remove hook

Pipeline Usage

# Strict rules check with JSON output
kodus review --rules-only --format json --fail-on error

# Generate markdown report artifact
kodus review --format markdown --output review-report.md

Authentication

Kodus supports multiple auth methods depending on your setup:

Trial Mode (no account)

Just run kodus review. No signup needed. You get 5 reviews/day with up to 10 files and 500 lines per file — enough to try it out. Sign up free to remove limits.

Personal Login

For individual developers. Creates a session with automatic token refresh.

kodus auth login           # Sign in with email/password
kodus auth status          # Check auth status and usage
kodus auth logout          # Sign out

Credentials are stored locally in ~/.kodus/credentials.json.

Team Key

For teams where not everyone needs their own account. A single shared key gives the whole team access — developers just set the key and start reviewing, no individual signup required.

kodus auth team-key --key kodus_xxxxx

Or set it as an environment variable:

export KODUS_TEAM_KEY=kodus_xxxxx

Get your team key at app.kodus.io/settings/cli. Team keys have configurable device limits managed from the dashboard.

This is also the recommended auth method for AI coding agents (Claude Code, Cursor, Codex) — set the env var once and every agent session is authenticated automatically.

CI/CD Token

For pipelines and automated environments. Generated from your personal login:

kodus auth token           # Generate a CI/CD token

Then use it in your pipeline:

export KODUS_TOKEN=<your-token>
kodus review --format json --fail-on error

Note: For PR-level reviews in CI/CD, we recommend using the Kodus platform GitHub/GitLab integration instead of the CLI. It's purpose-built for PR workflows with inline comments, status checks, and team dashboards.

Environment variables

Variable	Description
`KODUS_API_URL`	API endpoint (default: `https://api.kodus.io`). HTTPS only (except localhost).
`KODUS_TOKEN`	CI/CD token for automated pipelines (generated via `kodus auth token`)
`KODUS_TEAM_KEY`	Team key for shared team access and AI coding agents

Privacy & Security

Kodus sends your code diffs to the Kodus API for analysis. We take this seriously:

HTTPS only — All API communication is encrypted. Custom API URLs are validated.
No training on your code — Your code is not used to train models.
Minimal data — Only diffs and context files are sent, not your entire codebase.
Credentials stored locally — Auth tokens are kept in ~/.kodus/credentials.json on your machine.

Contributing

We welcome contributions! Please see our issues page to get started.

npm install       # Install dependencies
npm run build     # Build
npm run dev       # Watch mode
npm test          # Run tests

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 68 Commits
.claude/commands		.claude/commands
.github/workflows		.github/workflows
scripts		scripts
skills		skills
src		src
tests/integration		tests/integration
.gitignore		.gitignore
README.md		README.md
install.sh		install.sh
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json
vitest.config.ts		vitest.config.ts
vitest.integration.config.ts		vitest.integration.config.ts
yarn.lock		yarn.lock

kodustech/cli

Folders and files

Latest commit

History

Repository files navigation

Kodus CLI

Quick Start

What It Does

Code Review

PR Suggestions

Decision Memory

Best With AI Agents

Install the Skill (recommended)

How It Works With Agents

Setup: Claude Code

Setup: Cursor / Windsurf

Setup: Headless / Shared Environments

Copy & Paste Workflow (interactive)

Installation

Skill installer (recommended — CLI + all your agents)

CLI only

Review Modes

Interactive (default)

Auto-fix

AI Agent

Output Formats

Diff Targets

All Flags

Decision Memory

CI/CD & Git Hooks

Pre-push Hook

Pipeline Usage

Authentication

Trial Mode (no account)

Personal Login

Team Key

CI/CD Token

Privacy & Security

Contributing

License

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases 9

Packages 0

Uh oh!

Contributors 2

Uh oh!

Languages

Packages