GitHuman

Review AI agent code changes before commit.

GitHub revolutionized how humans collaborate on code.

GitHuman defines how humans review code written by AI.

The Problem

AI coding agents write code. But the traditional PR workflow assumes humans are the authors. By the time AI-generated code reaches a pull request, you've already committed to the approach. Review happens too late.

The Solution

GitHuman moves the review checkpoint to where it belongs: the staging area.

Before git commit, you get a proper review interface—not a wall of terminal diff output. Add comments, track issues, and make informed decisions about what the AI produced.

Screenshots

Reviews List

Staged Changes

Features

• Visual diff review - Review staged changes in a clean, GitHub-like interface
• Human handoff for agents - Run githuman ask to ask a human for review, wait, and hand their feedback back to the agent
• Inline comments - Add comments to specific lines with code suggestions
• Review workflow - Track status: in progress, approved, or changes requested
• Todo tracking - Create tasks for follow-up work via CLI or web interface
• Markdown export - Export reviews with comments for documentation
• Keyboard shortcuts - Navigate quickly with vim-style bindings
• Local & private - Everything runs on your machine, no data leaves

Requirements

• Node.js 24.0.0 or higher

Installation

npm install -g githuman

Or run directly:

npx githuman@latest serve

Quick Start

Review staged changes before commit

# Stage your changes (from AI agent or manual edits)
git add .

# Start the review interface
githuman serve

This opens a web interface at http://localhost:3847 where you can review your staged changes before committing.

Ask a human to review and wait for feedback

githuman ask "Please review the parser refactor"

githuman ask starts or reuses GitHuman, opens a dedicated ask page, waits for the human to finish reviewing, and then prints the new todos, comments, and review status updates for the agent.

The handoff is completed in the UI with a dedicated Continue assistant button.

Agent Skills

GitHuman provides an agent skill that teaches AI coding agents when and how to use GitHuman for reviewing changes.

Install it with the skills CLI:

npx skills add mcollina/githuman-skills

CLI Reference

Start Review Server

githuman serve [options]

Options:
  -p, --port <port>    Port to listen on (default: 3847)
  --host <host>        Host to bind to (default: localhost)
  --open               Auto-open browser
  --no-open            Don't open browser automatically
  --https              Enable HTTPS
  --no-https           Disable HTTPS
  --cert <path>        Use a custom TLS certificate
  --key <path>         Use a custom TLS private key
  --auth [token]       Enable auth, optionally with a provided token
  -v, --verbose        Enable verbose logging
  -h, --help           Show help

Ask a Human to Review

githuman ask [message] [options]

Options:
  -p, --port <port>    Port to use for GitHuman
  --host <host>        Host to bind to when starting GitHuman
  --open               Open the browser automatically
  --no-open            Don't open the browser automatically
  --review <id>        Scope feedback to a specific review
  --interval <ms>      Polling interval in milliseconds
  --json               Output final feedback as JSON
  -h, --help           Show help

List Reviews

githuman list [options]

Options:
  --status <status>    Filter by status (in_progress|approved|changes_requested)
  --json               Output as JSON
  -h, --help           Show help

Export Review

githuman export <review-id|last> [options]

Arguments:
  review-id            The ID of the review, or "last" for most recent

Options:
  -o, --output <file>  Output file path (default: stdout)
  --no-resolved        Exclude resolved comments
  --no-snippets        Exclude diff snippets
  -h, --help           Show help

Manage Todos

githuman todo <subcommand> [options]

Subcommands:
  add <content>     Add a new todo item
  list              List all todos (pending by default)
  done <id>         Mark todo as completed
  undone <id>       Mark todo as not completed
  remove <id>       Delete a todo
  clear --done      Remove all completed todos

Options:
  --review <id>     Scope todo to a specific review
  --all             Show all todos (not just pending)
  --done            Filter to show only completed todos
  --json            Output as JSON
  -h, --help        Show help

Workflow

Staging-area review

1. AI agent makes changes - Claude, Copilot, Cursor, or any tool stages code
2. Run githuman serve - Opens the review interface
3. Review the diff - See exactly what changed, file by file
4. Add comments - Note issues, questions, or suggestions
5. Create todos - Track follow-up work
6. Decide - Approve and commit, or request changes from the agent
7. Export - Optionally save the review as documentation

Human-in-the-loop agent handoff

1. The agent runs githuman ask
2. GitHuman starts or reconnects and prints the review URL
3. The human reviews in the browser and leaves comments or todos
4. The human clicks Continue assistant when feedback is ready
5. githuman ask exits with the new feedback so the agent can continue

Web Interface

Creating a Review

1. Stage your changes with git add
2. Run githuman serve
3. Click "New Review" or navigate to Staged Changes
4. Click "Create Review"

Adding Comments

1. Hover over any line in the diff
2. Click the + button that appears
3. Write your comment
4. Optionally add a code suggestion
5. Click "Add Comment"

Keyboard Shortcuts

| Key | Action | |-----|--------| | j | Next file | | k | Previous file | | Esc | Cancel / Close |

API

The server exposes a REST API with OpenAPI documentation at /docs.

Authentication

Set a token to require authentication:

# Via CLI flag
githuman serve --auth mysecrettoken-that-is-at-least-32-characters

# Auto-generate a token
githuman serve --auth

# Via environment variable
GITHUMAN_TOKEN=mysecrettoken-that-is-at-least-32-characters githuman serve

Clients must include the token in the Authorization header:

Authorization: Bearer mysecrettoken-that-is-at-least-32-characters

Configuration

GitHuman can read repo-local defaults from .githuman/config.json.

{
  "host": "localhost",
  "port": 3847,
  "https": false,
  "open": true
}

Config precedence is:

1. CLI flags
2. .githuman/config.json
3. built-in defaults

Data Storage

GitHuman stores local data in:

<repository>/.githuman/

Important files include:

• reviews.db - reviews, comments, and todos
• config.json - optional repo-local CLI defaults

This directory is typically gitignored.

Development

# Clone the repository
git clone https://github.com/mcollina/local-code-reviewer.git
cd local-code-reviewer

# Install dependencies
npm install

# Run server in watch mode
npm run dev:server

# Run web dev server (Vite)
npm run dev

# Run all tests
npm test

Tech Stack

• Backend: Fastify, Node.js native SQLite
• Frontend: React 19, Vite, Tailwind CSS v4
• Testing: Node.js test runner, Vitest, Playwright

Why "GitHuman"?

In the age of AI coding assistants, someone needs to review the code before it's committed. That someone is you. GitHuman is the human checkpoint in an AI-assisted workflow.

License

MIT License for the code.

The GitHuman logo is Copyright (c) Matteo Collina, All Rights Reserved.