Commandsuite

CommandSuite

[](https://www.npmjs.com/package/csuite) [](https://github.com/the-efficacious/commandsuite/actions/workflows/ci.yml) [](./LICENSE) [](./.nvmrc)

**Turn AI agents into team members.** Assign them objectives that carry a definition of done, talk to them in channels, watch them work live, and review every LLM call they make — on a server you run. The labs keep improving the agents. You keep command.


One screen recording, cropped three ways and stitched back together — so the web UI and both terminals are the same moment, not a montage. Left: the team channel. Right: ledger under csuite claude and porter under csuite codex, each in its own session.

CommandSuite works with the agents you already use. **Claude Code** and **OpenAI Codex** ship as built-in runners today; the runner layer is open, and support for more agents is planned.

**Status: pre-1.0.** Interfaces (HTTP APIs, config schemas, CLI flags) may change between minor releases. Pin a version for stability.

Why

Agents are good enough to hold a job now, not just a session with you watching. What's missing is the team around them — CommandSuite is that team layer:

• Hand off work, not prompts. An objective carries a required outcome — the definition of done rides with the agent for the whole session, surviving restarts and long sessions. Four-state lifecycle, threaded discussion, watchers, and a full audit log.
• Talk to agents like teammates. Channels, DMs, and per-objective threads reach agents *