Skip to content

quickstart.md

Latest commit

 

History

History
66 lines (40 loc) · 5.23 KB

quickstart.md

File metadata and controls

66 lines (40 loc) · 5.23 KB
Raw

Quickstart

For developers who want to get a team of coding agents running on a real repo in under 10 minutes.

By the end of this guide you will have AgentsCommander installed, an AC project pointing at one of your repos, a Team with two agents, and an orchestrator that delegates a task to a worker through a markdown message.

Prerequisites

  • A host covered by the platform support policy. Windows 10 1809+ or Windows 11 on x86_64/AMD64 is fully supported. Linux x86_64/AMD64 is supported with medium test coverage — expect some turbulence — and requires explicit confirmation before installation. macOS is supported via npm only, with very low test coverage, so expect errors.
  • At least one supported coding-agent CLI (Claude Code, Codex, Antigravity, or Pi) installed and authenticated as its CLI requires. See Installing the coding-agent CLIs for the exact upstream links and Pi install commands. You can install more than one; AC lets you pick per agent.
  • Node.js 18+ and npm only if you choose the npm route; on macOS it is the only supported route.
  • Git installed.
  • A repo you want the agents to work on (it can be empty).

You do not need Rust to run AgentsCommander. Rust is only needed if you want to build from source — see CONTRIBUTING.md.

1. Install AgentsCommander

Use the copyable install-with-a-Coding-Agent prompt. The Coding Agent reads the pinned installation contract, reports your host's support tier, selected stable release asset, checksum, destination, commands, privilege and PATH effects, validation, and rollback, then waits for your approval.

After it verifies the exact asset against SHASUMS256.txt and validates the executable, start AgentsCommander from the approved location. If you choose a manual release asset or npm, use the same platform and trust gates in the canonical guide.

2. Open or create an AC project

An AC project is a folder with a Project AC Root (.ac/) inside. AC stores agents, teams, and messaging here so the whole project is portable and version-controllable. The repository that versions .ac/ is the Agents config repo: giving .ac/ a repository of its own is the recommended layout, and tracking it inside a work repo also works.

In the sidebar, click New Project and point at any folder, empty or an existing repo. AC creates .ac/ with a sensible .gitignore and registers the project in your sidebar.

CLI equivalent: agentscommander new-project /path/to/folder

3. Create your first Team

Open the Teams pane. A Team is one orchestrator plus one or more worker agents working toward a shared goal. The agents are created first, then the team definition links those existing agents together.

  1. Click + Team and give it a name (for example feature-x).
  2. Add the orchestrator — give the agent a directory name (for example tech-lead), pick a role template from the Agents Agency picker, and finish.
  3. Add one worker the same way (for example dev-rust with the Rust developer template).
  4. Mark the first agent as orchestrator.

Behind the scenes AC creates agent matrices under .ac/_agent_tech-lead/ and .ac/_agent_dev-rust/, then saves the team definition under .ac/_team_<team-name>/.

4. Write a brief and launch the orchestrator

Activate the team for a task and give the room a title. AC creates .ac/room-1-<team-name>/, including TASK.md, messaging/, and each __agent_<name>/ replica. Open the new TASK.md and write what you want the team to do. One paragraph is enough; the orchestrator will expand it.

Click the orchestrator's session in the sidebar. AC opens a real terminal and prompts you to pick a coding agent (Claude Code, Codex, Antigravity, or Pi). Pick one. AC launches the agent in the orchestrator's directory with the role and brief already loaded.

5. Watch the agents exchange a message

Ask the orchestrator something like:

"Send a hello message to <project>:room-1-feature-x/dev-rust and ask them to confirm the role they have."

The orchestrator will write a markdown file to .ac/room-1-feature-x/messaging/ and run agentscommander send --to <peer> --send <filename> --mode wake. In a second or two you will see the worker's session activate, read the file, and reply by writing its own message back.

That's the loop. Every message is a file you can cat, git diff, and audit.

Next steps