The loop
Stop
/pitwall:stopEnd the loop after the task in progress finishes.
What it does
/pitwall:stop ends the loop gently. A task in progress is left alone: its round finishes, opens the PR, and then the loop stops by itself. Between rounds, it stops at once.
Use it when
- you want the loop to end but the current task to land as a PR
- you are back and want the checkout for yourself after this task
To stop in the middle of a task, use /pitwall:brake.
Try it
/pitwall:stop
What you’ll see
The examples come from a made-up repo, acme/shop. A task is in progress:
Will stop after P2-05 finishes (currently at step reviewing).
Between rounds, a summary of the run:
Loop stopped
Done
P2-04 #58 PARTIAL
P2-07 #59 VERIFIED
Parked
P2-05 agent/p2-05-saved-card the plan does not say what happens when the saved card has expired
When it stops, it also reads the session for anything you keep repeating, and speaks only if there is something new.
The playbook the agent followsThe full spec for /pitwall:stop. This page sums it up; when the two differ, the playbook wins.
/pitwall:stop — stop after the current task
- Read state (path in
.claude/agent-loop.md; no file + the profile has a “multi-machine” section → receive first) — none /running:false→ say “the agent is not running” (in the profile’s user language), end - Set
stopRequested: trueand write it back (multi-machine → send state every time you write) currentTaskis null (between rounds / queue empty / waiting for merge) →running:false,ScheduleWakeup stop:true, summary report ofdone(PR link + gauge status) andparked(branch + reason + question)- Has
currentTask→ don’t touch the task, tell the user: “will stop after<id>finishes (currently at step<step>)” — the current/pitwall:lapround works until the PR is opened, then stops itself - Want to stop right now → tell them to use
/pitwall:brake - At the end of the item 3 report →
/pitwall:secretaryshort mode (speak only if there is something new that keeps being repeated)