feat(tmux-agent-orchestrate-monitor): integrate watchdog pattern as skill

Moved /tmp/subscriber-watchdog.sh → skills/tmux-agent-orchestrate-monitor/scripts/watchdog.sh
(skill-managed lifecycle, no longer lives outside workspace).

Added lib.sh::start_watchdog() helper:
- Spawns watchdog as background nohup process
- Writes watchdog log to .hermes/jobs/<JID>.watchdog.log
- Returns watchdog PID via stdout

Wired create_session.sh --submit-job to auto-start watchdog after JOB registration.

Fixes:
- Bug: registry.py get first-line parse was fragile (empty status → infinite loop)
  → Now uses python3 json.load for robust parsing
- Bug: old path skills/delegate-job/scripts/job_subscriber.py hardcoded
  → Now uses skills/tmux-agent-orchestrate-delegate-job/scripts/job_subscriber.py

Verified on isolated server -L agy-watchdog-skill-test (kill-server after):
- Syntax check PASS
- E2E: register job → start watchdog → publish completed → watchdog exits
- Global skill non-interference verified
- Main isolated server -L multi-agent-canary untouched

skills/lib.sh

@@ -425,3 +425,25 @@ delegate_publish_event() {
   "$py_bin" "$pub" --job "$job_id" --event "$event" --detail "$detail" || true
 }
 
+# start_watchdog <job_id> [workdir]
+# Spawns a watchdog process to monitor a delegate-job JOB in the background.
+# The watchdog re-spawns the subscriber every 2 minutes (or whatever hard
+# limit we set) and exits automatically when the JOB reaches terminal state.
+# Returns the watchdog PID via stdout.
+start_watchdog() {
+  local job_id="$1"
+  local workdir="${2:-$PWD}"
+  local watchdog_script="$workdir/skills/tmux-agent-orchestrate-monitor/scripts/watchdog.sh"
+  local log_file="$workdir/.hermes/jobs/${job_id}.watchdog.log"
+  
+  if [ ! -x "$watchdog_script" ]; then
+    echo "ERROR: watchdog not found or not executable: $watchdog_script" >&2
+    return 1
+  fi
+  
+  nohup "$watchdog_script" "$job_id" "$workdir" > "$log_file" 2>&1 &
+  local pid=$!
+  echo "$pid"
+}
+
+