resume-runlisted

# Role Convenience wrapper. Show progress, then advance the pipeline. # Invocation /ultra-analyzer:resume-run [run-name] # Protocol ## Step 1: Call /ultra-analyzer:progress Invoke the progress skill to print current state. User sees where they paused. ## Step 1b: Auto-heal orphan mkdir-locks (closes H-6) EXIT traps in claim.sh / state.sh inc / state.sh dec / state.sh set / requeue.sh do NOT fire on `kill -9` or power loss. The mkdir-based lock dirs persist as orphans and every subsequent `state.sh inc/set/dec` then spins for 30s before failing. `/resume-run` MUST detect and clear orphan locks BEFORE handing off to /run. For each candidate lockdir: - `<RUN_PATH>/topics/.claim.lock.d` - `<RUN_PATH>/state.json.lock.d` Apply the heuristic: ```bash heal_orphan_lock() { local lockdir="$1" [[ -d "$lockdir" ]] || return 0 # not held; nothing to do # If the lockdir owner wrote a PID file, prefer that. local pidfile="$lockdir/holder.pid" if [[ -f "$pidfile" ]]; then local pid pid=$(cat "$pidfile" 2>/dev/null) if [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null; then echo "[resume-run] lock $lockdir held by live PID $pid — leaving alone" >&2 return 0 fi fi # Otherwise fall back on stat-mtime: lockdir older than 30s with no live # holder is presumed orphaned (kill -9 / power loss). local age_s if stat -f '%m' "$lockdir" >/dev/null 2>&1; then # macOS / BSD stat local mtime; mtime=$(stat -f '%m' "$lockdir") age_s=$(( $(d