Repository navigation
make a way to easily see in progress thirdshift runs on machine and cut one off #574
Description
such as if a pickup run started on an issue before I was ready to have that issue picked up and started on
Concrete example: Chart35 #239, “Spec: Couple Sharing for two accounts with one chart history.” On October 8, 2026, a cron-triggered thirdshift pickup base main took it and started a Codex implementation session before I considered it ready. The activity log recorded the run starting at 9:17:08 a.m. EDT.
I asked Codex to find and stop it, and also asked how I could see and stop runs myself. I was unsure whether it was running on another server. Codex found it on the server whose hostname is thirdshift, by inspecting processes and the repository’s activity and command logs. There was no built-in run-listing or stop command in the installed CLI.
To stop it, Codex sent SIGTERM to the Thirdshift Pickup parent process (kill -TERM 1826215), rather than killing the Codex process directly. Thirdshift stopped the session, pushed the issue-239 branch during cleanup, removed the local worktree and branch, and recorded Pickup run #239 ended: failed: interrupted at 9:22:16 a.m. EDT. The process tree exited; no PR was created.
This case illustrates the need to identify which issue each active run is working on, which machine it is on, and its logs, then stop the selected run gracefully and see confirmation that it stopped, including any branch or PR it left behind.
Activity
JacobStephens2 commentedon Oct 17, 2026
Triage Notes
#711 now owns viewing active Runs across machines; this issue still owns selecting one and stopping it gracefully. Existing SIGTERM handling can preserve work, but there is no supported discovery/stop command.
Decisions needed: How should a Run be selected by issue and machine, and is the first version local-only? Define authorization/transport for a remote stop, confirmation that the selected process actually ended, and the branch/PR information shown afterward. Coordinate Claim handback with #613/#617 rather than assuming a stop automatically makes the issue eligible for Pickup.
JacobStephens2 commentedon Oct 21, 2026
Closing as already implemented, per machine:
thirdshift statuslists this machine's live Commands with their issue, stage and age.--jsongives each one's Command log and Session log, andthirdshift logsreads them.thirdshift stop <command>sends SIGTERM to the recorded process only, waits for its cleanup, and prints the outcome and the Issue branch and pull request it left. The Console's Stop does the same after a confirmation page.Related follow-ups: stopping one Ticket's Run inside a Spec run (#832), and saying which machine holds a Claim (#833).