diff --git a/.kilo_workflow/learnings/e2e-slot-owner-must-be-dedicated-session.md b/.kilo_workflow/learnings/e2e-slot-owner-must-be-dedicated-session.md new file mode 100644 index 0000000000..b50a52cc2f --- /dev/null +++ b/.kilo_workflow/learnings/e2e-slot-owner-must-be-dedicated-session.md @@ -0,0 +1,17 @@ +# E2E slot acquired under a shared/long-lived tmux session name leaks the slot + +Symptom: an E2E slot stays held for tens of minutes with no verifier running; `e2e-slot.sh status` +shows the holder as a long-lived session name (e.g. the planner/starter session) while other +sections starve on a 3-slot machine. The slot is never auto-reclaimed because the holding session +never dies. + +Cause: an orchestrator (not a verifier) ran `e2e-slot.sh acquire` itself, under its window's shared +tmux session name instead of a dedicated per-round verifier session. Slot reclamation keys on the +holder's tmux session being gone; a shared session outlives every round, so the hold is effectively +permanent until someone releases it by hand. + +Fix: only the e2e-verifier acquires a slot, always under its own dedicated session name +(`
-e2e-verifier-