Runtimes
Agent Teams drives the agent CLIs you already have, on your own subscription or key. It does not resell model capacity and does not meter your usage.
| Runtime | Use it for | Why |
|---|---|---|
| Claude Code | leads and reports | Takes Agent Teams' own tools (delegate, await results, report, ask), and a per-run tool list |
| goose | leads and reports | Takes Agent Teams' tools, and is asked about every tool call, so the agent's mode is enforced per call |
| Codex | review, or one self-contained task | Takes Agent Teams' own tools, but no per-run tool list, so its worktree and the gate are its limits |
lucitra runtimes # what is installed, and which one your chief of staff answers on
lucitra runtimes goose # switch
lucitra configure devops runtime codex # one agent on its own runtime
lucitra models # which model each agent runs, on which runtime
A task runs on the runtime you set on its agent with configure, and otherwise on the one it was
started from: the session's for work you start, the lead's for work a lead delegates. So a Claude
Code lead with its reviewer set to codex has Codex review.
Modes
Each agent has a ceiling: plan, manual, accept-edits or auto. No agent reaches merge.
accept-edits writes and commits inside its worktree and stops before pushing. On goose that
ceiling is checked at every tool call; on Claude Code it is the tool list the run starts with.
From another agent
lucitra mcp
Serves Agent Teams' commands over MCP, so an agent you already use can start a task, read the task list or answer a gate. It is the same command list the shell and the app use, with the same checks.