Playbook: The Driver
The playbook for humans who want the suite to keep going by itself. /continuous-refactoring is the Driver: it works one round after another, with a cadence between them, until an end. You are asked only where a decision is yours.
A round is one run, the review of its merge request, and any rework:
check the project → a run in a subagent → review its merge request → settled, or rework → wait the cadence → next round
The call
It starts without asking, and says back in one sentence what it took. Everything has a default; say otherwise in your own words.
| Setting | Default | You can say |
|---|---|---|
| Merge | review only: it comments, and you merge |
merge |
| End | five rounds | a number of rounds, of merged merge requests, a time, "until nothing is workable" |
| Cadence | 30 minutes between rounds | another duration, "no pause" |
| Call | none: each run makes its own Track choice, Investigation included | a Track ("Guardrails"), a restriction ("only the Rector tools") |
| Log | none | a file your repository ignores, one line per round. The log is for you; the Driver reads its state from the tracker |
For one run alone, a named ticket or a Pack of your own, call /continuous-refactoring-run. Given one of those, the Driver passes it on.
What a round does
- Every run starts fresh. A run in a subagent has a context of its own, so the hundredth run reads the tracker as clearly as the first. It is always autonomous, and your
AGENTS.mdrules go with it. - It names a ticket to go on with one: to rework a merge request it or you asked changes on, for the next part of a ticket worked in several merge requests, for a question you answered, or when you ask. It never names a ticket that waits or is blocked: naming it would work it whatever its state.
- It reviews every merge request in a subagent of its own, with your project's review skill where there is one, otherwise as an engaged human reviewer would: in scope of its ticket, behaviour kept, checks green, the description says why the change is safe.
- Rework stays in the round. A request for changes is posted on the merge request, and a run answers it at once, with no cadence between. After two reworks without a mergeable result the merge request goes to you. A round holds up to three runs, and counts as one.
- With
merge, it waits for green CI and uses your repository's merge method. It checks for a merge request stacked on the branch before deleting it, and that the merge really landed the branch's content. If not, it stops: the correction is a normal merge request, never a rewrite of history.
The cadence
It is counted from the end of the round before. The Driver waits in its session where your environment has a timer. Otherwise it ends and says when the next round is due, and the next call works the next round. A short cadence is all a session holds; a longer one, once a day say, is yours to set up with your own scheduler. See known limitations.
Where you come in
The Driver never answers an open question itself, and it goes on with the next ticket while one waits for you.
| What waits for you | Where it is | What you do |
|---|---|---|
| A question that came up while planning | an Open question comment on the ticket, posted by the run |
answer on the ticket; the Driver names it once you have |
| A question the Driver has about a merge request | a comment on the merge request | answer there, and call again |
| A merge request reworked twice without passing | a comment on the merge request | as above |
A mergeable merge request, with review only |
a comment on the merge request | merge it, or ask for changes there: the next round reworks it |
A change a run wants in your AGENTS.md |
the Driver's message, with the wording | make it yourself, or not |
It tells you about each as it happens, through whatever notification your environment has, and writes one status line after every round. It leaves comments on waiting tickets alone: any comment there would count as your answer.
How it ends
- Done: a run finds nothing workable, and nothing waits.
- Paused: a run finds nothing workable but tickets wait, are in review or are blocked; or, with
review only, more than three merge requests are open. It lists what each needs. Call again once you have dealt with them. - Something broke: a run failed, the working tree is not clean after a run, a merge did not land, or two runs in a row ended without a merge request and without a new waiting ticket.
- Your end: by default the fifth round.
Who starts it
You or a schedule, by its name. An agent that wants refactoring done calls /continuous-refactoring-run; a Driver started in a subagent stops and says so.
The first call
On a project without a ## Refactoring operations section the onboarding asks you questions, and a run in a subagent can ask nobody. So the first call runs the onboarding right in your session, and stops: commit what it wrote, then call again.