Introduction

repo.box is a git permission layer that makes repositories safe for AI agents.

The Problem

GitHub's identity model was designed for humans. One account, one set of permissions, one audit trail. When an AI agent pushes code, it borrows a human's token and inherits all their access. There's no way to scope an agent to specific files, branches, or actions.

This creates a trust bottleneck: you can't give an agent enough access to be useful without giving it enough access to be dangerous.

What repo.box Does

repo.box sits between your agents and your git repos. It provides:

• Cryptographic identity for agents (EVM keypairs, not borrowed tokens)
• Declarative permissions defined in .repobox/config.yml (file paths, branch patterns, PR limits)
• Group-based access (define teams of humans and agents, assign permissions to groups)
• Workflow state machines (PRs, reviews, releases as configurable pipelines)
• Sandboxed experimentation (agents can go wild in their branches; production paths are gated)

How It Works

1. Every participant (human or agent) has an EVM identity
2. The repo owner defines rules in .repobox/config.yml
3. The repobox CLI enforces rules locally on commit/push
4. The server enforces rules again on receive (no bypass possible)
5. All actions are signed and auditable

Quick Start

# Install the CLI
cargo install repobox

# Initialize a repo
repobox init

# Add an agent identity
repobox identity add evm:0x1234...abcd --alias ocean

# Define permissions in .repobox/config.yml
# Push with enforcement
git push

See the Identity spec to understand the foundation layer.