daily-fix: move tmux sockets off /tmp (split-brain)

Overview / Motivation

Auto-filed by /daily 2026-07-16 from transcript mining (two chats hit it independently). Non-workflow-surface (system/shell config + cron wrappers). Root-cause identification is part of the task: rule out the #911 /tmp sweep as the deleter before excluding paths.

Goal

Make tmux server sockets survive /tmp cleaning on the long-uptime VM, and identify what deleted the socket.

Workflow gap

• Bug observed: the tmux server socket /tmp/tmux-1001/default was deleted on the 116-day-uptime VM (root cause unidentified — check systemd-tmpfiles age rules and the #911 non-canonical /tmp sweep's match set), so the 06:00 mygoat cron silently started a SECOND tmux server and 39 sessions became invisible to tmux ls (user report ~17:50-18:00Z, chats c07f01a6 + 108c810d); recovered by rebinding the old socket as /tmp/tmux-1001/old
• Why it is a workflow gap: The program orchestrator, mygoat crons, and manual sessions all assume one tmux server; a /tmp-cleaned socket silently forks the world.
• Confidence (emitter): high (incident today; recovery confirmed; deleter unidentified)
• verified-at-filing: incident + recovery in chats c07f01a6/108c810d (39 sessions on the rebound socket); n/a for a repo grep — target is system config + cron wrappers (grep -rln tmux scripts/cron_*.sh enumerates wrappers at planning time)

Proposed change (candidate diff sketch — refine in planning)

export TMUX_TMPDIR=$HOME/.tmux-sockets in the shell profile AND every cron wrapper that touches tmux; identify + exclude the /tmp cleaner that removed the socket; document the recovery

Scope / surfaces

• Primary target: shell profile + scripts/cron_*.sh
• Grep the workflow surface for the pattern before editing and update every hit; list them in the plan.

Constraints / invariants

• scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
• Non-workflow-surface fix (wf_fix: false): no recursion guard applies; standard /issue pipeline.