Skip to main content

Running a team

mmt run <team> "your task"

It runs on your subscription​

mmt run auto-opens an interactive Claude Code session and executes the team's members as sub-agents inside it. That means it uses your Claude subscription, not claude -p / the Agent SDK credit.

  • Inside a Claude Code session you can also run it directly: /mmt run <team> "task".
  • From a plain terminal, mmt run … launches that session for you (one command).

Choose the model​

mmt run spec-to-prod "…" --model opus        # force one model for the whole run

Precedence: --model → member model: → team model: → your default. Set per-member models for cost/speed (e.g. reviewer on sonnet, coder on opus).

Preview for free​

mmt run spec-to-prod "…" --sim

A simulated tracker with no Claude calls — handy to check the shape and timing of a workflow.

Watch it work (observability)​

While a team runs you see, live:

  • who is active — the team lead vs a specific member,
  • elapsed time total and per member,
  • the loop round it's on,
  • steps that are pending, done, or skipped.

Every real run also writes a full report to ~/.my-mini-team/runs/<team>-<timestamp>.md.

Opt-in run audit on the PR​

Add a reporter member (plugging the publish-report skill) or set report: github on a team, and a run posts a concise, collapsible audit — steps, per-member time, rounds, verdicts, total, and a link to the full report — as a comment on the PR. It's strictly opt-in: teams without it post nothing.

Off-subscription escape hatch​

--headless runs via claude -p instead. This draws from the Agent SDK credit / API, not your subscription — avoid it unless you specifically want headless/automated execution.