Agentlab — Productivity skill for Claude Code

AgentLab

[](https://github.com/RiadMefti/agentlab/actions/workflows/ci.yml) [](https://github.com/RiadMefti/agentlab/actions/workflows/release.yml)

**One captain per project folder, with all your coding agents in one fast local terminal.**

AgentLab runs Codex, Claude Code, and OpenCode as their real CLIs. They keep their existing authentication, configuration, tools, and context. Each saved project is a named local folder with exactly one captain. The captain may coordinate any number of workers, and every live session remains directly accessible.

┌─ PROJECTS ───────┬─ SELECTED AGENT TERMINAL ───────────────┬─ AGENTS ─────────┐
│ named folders    │ full ANSI/PTY interaction               │ CAPTAIN (pinned) │
│                  │                                         │ WORKERS           │
└──────────────────┴─────────────────────────────────────────┴───────────────────┘

There is no web server, browser UI, Electron shell, or remote mode. The application is a single-process terminal compositor that calls the local application layer directly. tmux owns the durable sessions; the UI attaches exactly one PTY to the selected agent.

Install

Install the single public package, then start the app from anywhere:

npm install --global agentlab
agentlab

The npm package downloads the matching AgentLab executable from GitHub on first use, shows live download progress, and caches it by version. Later launches use the cached executable and make no update request. Updates are always explicit:

agentlab update --check
agentlab update

You can alternatively download an executable from the [latest GitHub release](https://github.com/RiadMefti/agentlab/releases/latest):

• agentlab-vVERSION-linux-x64
• agentlab-vVERSION-mac-arm64

Then make it executable and place it on y