Two tasks. Two routes.
A typo fix that closes on its own, and a payment change that stops twice for you. Watch the scope guard refuse an edit, the budget count the suite runs and the dialog carry your decisions.
Route
Red dots are the stops that wait for you.
/rr-status
An illustrative session. The outputs follow docs/getting-started.md and the formats state.py prints; task ids, hashes and file names are examples.
The parts a model does not decide.
The weight
Seven lines before any edit. In the payment task, src/Payment/** is in the project's money modules, so domain is 2 and the task is full even though W is only 6.
The scope refusal
The typo task tries a second template. The hook refuses it and names the step. The agent either drops the edit or widens the task on record with a reason.
The budget
The payment suite goes red once. The failure is fixed as one batch, and the second run is the last one the agent gets. A third run would be your call.
The dialog
The question, the plan, the approval and the commit come to you in the runtime's own dialog. What you pick is recorded by a hook and runs once, verbatim.
| You want to | Run |
|---|---|
set up .ai/, or bring it up to the installed plugin | /rr-setup, /rr-setup --update |
| score the repository against the twelve plays | /rr-setup --audit |
| turn an idea into intent, requirements and design | /rr-spec <idea> |
| plan a spec into tasks, or one task into a mode | /rr-plan |
| do the work, or resume it | /rr-build <request> |
| run a whole plan as one task | /rr-build-all <plan> |
| review the change | /rr-review |
| close the task, get your approval | /rr-ship |
| see where it stands, or what it cost | /rr-status, /rr-status --cost |
4.0 removed the 3.x alias names; only the rr-* commands remain.