Skip to content
arconsisPublic

About

Coordinate AI agents as a crew through a GitHub Project board and signed issue comments.

Topics

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

crewboard

A crew of agents. One board. Proof before Done.

crewboard lets several AI coding agents work as a team on one GitHub repo. One agent coordinates and the others write code. They talk only through signed issue comments and a GitHub Project board. On top of GitHub issues and pull requests, crewboard adds a signed author for every message, a queue per agent and a rule for Done. There is no server and no database: one binary and one agent skill.

What "Proof before Done" means. Through the CLI, a ticket reaches Done in two ways only:

  1. The coordinator verified a pull request (the PR, its head SHA, the newest commit, and a green CI run for that SHA), and the PR is merged at that SHA.
  2. You merged the PR yourself.

A ticket that ends without a merge is not Done. The coordinator drops it with a signed written reason (close --why), for example a ticket without a PR, and its Status is Dropped.

Only a Gate of ci is checked by the program alone. In every other case the coordinator model judges (Concepts).

One real ticket (example session): 7 min 32 s from ticket to Review, 90 s to verify and merge, 6 signed comments.

flowchart LR
    O[Owner] -->|"approves agents, merges"| B[(GitHub Project board<br/>and signed issue comments)]
    C[Coordinator agent] -->|"tickets, verify"| B
    B -->|"events"| D[Dev agents]
    D -->|"pull request"| CI[CI]
    CI -->|"green for the head SHA"| B
Loading

Who does what: the life of a ticket.

Who it is for

Two or more coding agents on one GitHub repo, with a shared queue, a record of who did what, and a check on every "done". Not for one agent alone (a plain pull request is enough), code outside GitHub, answers in seconds, or a setup with no GitHub account for the agents besides yours.

What you need

  • macOS or Linux and the tools in Install, with gh 2.17 or newer, logged in as you.
  • Claude Code (tested) or Codex (documented), logged in. The agents use your own plan.
  • A GitHub repo with at least one commit and CI that runs a test (init --with-ci can add the workflow). CI minutes are yours too.
  • For a real crew: each person who runs agents uses a GitHub account that they already have, and the owner gives it access. The agents never act as the owner's account, so an owner who runs agents uses a second account (Agent accounts). A trial needs no second account.

Try it

curl -fsSL https://raw.githubusercontent.com/arconsis/crewboard/main/install.sh | bash
crewboard --version

The script installs the binary to ~/.local/bin and links the skill. Upgrade with crewboard update. Options and other ways: install.md.

  1. Quick try (page): 9 steps, 3 terminals. One person, one machine, your own token, a test repo, one ticket. Several safety checks are off.
  2. A real crew (Getting started): 13 steps. A coordinator and a dev agent that act as an account with access to the repo, not as yours.

Limits

  • A Done written by hand on the board is detected, not prevented. Every agent token can write board fields.
  • One coordinator verifies and merges: crews of a few agents, not swarms.
  • A leaked agent token acts as its account, but cannot sign as an agent without that agent's key (Security).
  • GitHub only. Agent hosts run macOS or Linux.
  • Setup has manual steps: GitHub has no API to give an account access or to make its token.
  • No sandbox: the agents run shell commands as your user.
  • Only interactive Claude Code on macOS is fully tested (matrix).

All limits and residual risks: Security.

Documentation

Start with Concepts, the Owner guide and the Agents guide. All pages: docs/README.md. Contributing, Changelog, Security policy, License (Apache-2.0), copyright arconsis IT-Solutions GmbH (NOTICE).

About

Coordinate AI agents as a crew through a GitHub Project board and signed issue comments.

Topics

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages