repo.box / git

Git permission layer that makes repositories safe for AI agents.

EVM keys, a git shim, and a .repobox/config.yml file that stands between your agents and your codebase. One config file. Per-agent permissions. Enforced before anything touches your repo.

$curl -sSf https://repo.box/install.sh | sh
copy
▶ Try in Playground
↓ scroll
Loading status...
Loading activity data...

What repo.box Does

Permission Layer
git shim
Intercepts git commands. Every commit, merge, push is checked against .repobox/config.yml before it happens.
EVM Identities
crypto
Agents sign commits with secp256k1 keys. No SSH keys, no GPG complexity. Just wallet addresses.
Sandbox by Default
security
Agents get feature branches. They can't touch main, can't edit the config. Structurally impossible to escalate.

Minimal Config

# .repobox/config.yml — protect main
groups:
  founders: [alice]
  agents:   [codex-bot]

rules:
  - group: founders
    allow: ["*"]

  - group: agents
    allow: ["feature/**"]
    deny:  ["main", ".repobox/*"]

Drop this in your repo. Agents get feature branches, founders keep full control.

Design Principles

  • →Agents use normal git — the shim is invisible
  • →EVM keypairs as the identity layer
  • →Every agent signs commits with its own key
  • →Everything is a file in the repo — rules, identities, workflows
  • →Enforced locally, no server required

Get started

One command. Secure your first repo in under a minute.

$curl -sSf https://repo.box/install.sh | sh
copy
▶ Try in Playground📄 SKILL.md for agents🤖 llms.txt