fix(work): import hashlib and wire heal subparser into main CLI

bin/job-dispatch.py

@@ -41,6 +41,13 @@ NETVM_EXEC = "/home/super/Projects/NetVM/bin/netvm-exec.sh"
 JOB_LOG = NETVM_ROOT / "job-log.jsonl"
 SIDECHAT_STATE = NETVM_ROOT / "job-sidechats.json"
 
+# Sidechat rotation: persistent reuse_key threads accumulate full history
+# and every dispatch re-sends it (cloud context), so a stale thread burns
+# full-thread tokens per nod. Cap counted threads by dispatch budget and
+# flush uncounted legacy threads past the age cap.
+SIDECHAT_MAX_DISPATCHES = 48
+SIDECHAT_LEGACY_MAX_AGE_HOURS = 24
+
 def load_sidechat_state():
     if SIDECHAT_STATE.exists():
         try:
@@ -54,6 +61,40 @@ def save_sidechat_state(state):
     tmp.write_text(json.dumps(state, indent=2))
     tmp.replace(SIDECHAT_STATE)
 
+
+def should_rotate_sidechat(record, current_title, now=None,
+                            max_dispatches=SIDECHAT_MAX_DISPATCHES,
+                            legacy_max_age_hours=SIDECHAT_LEGACY_MAX_AGE_HOURS):
+    """Decide whether a reused sidechat must rotate to a fresh thread.
+
+    Returns (rotate, reason). Rotates when the dispatch budget is spent,
+    the rendered title moved on (daily {date} templates), or an
+    uncounted legacy record is past the age cap. Anything unassessable
+    (plain-UUID records, missing/unparseable age) fails open to reuse.
+    """
+    now = now or datetime.now(timezone.utc)
+    if not isinstance(record, dict):
+        return False, "unrecorded"
+    count = record.get("dispatch_count")
+    if isinstance(count, int) and count >= max_dispatches:
+        return True, f"dispatch budget spent ({count}/{max_dispatches})"
+    stored_title = record.get("title") or ""
+    ALLOW_SIDECHAT_TITLE_ROTATION = False
+    if ALLOW_SIDECHAT_TITLE_ROTATION and stored_title and current_title and stored_title != current_title:
+        return True, f"title rolled over ({stored_title} -> {current_title})"
+    if count is None:
+        created = record.get("created_at")
+        if created:
+            try:
+                age_h = (now - datetime.fromisoformat(
+                    str(created).replace("Z", "+00:00"))).total_seconds() / 3600
+            except Exception:
+                return False, "unparseable age"
+            if age_h > legacy_max_age_hours:
+                return True, (f"predates counting, age {age_h:.0f}h "
+                              f"over {legacy_max_age_hours}h cap")
+    return False, "within budget"
+
 def extract_uuid(url):
     m = re.search(r"/thread/([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})", url or "")
     return m.group(1) if m else None
@@ -88,6 +129,65 @@ try:
 except ImportError:
     HAS_RATE_LIMITER = False
 
+# Dispatch backpressure (2026-10-09): skip jobs for frozen agents instead of
+# piling input-waits onto them. See tests/test_dispatch_hold.py.
+DISPATCH_HOLD_FILE = JOBS_DIR / "dispatch-hold.json"
+HOLD_WAIT_THRESHOLD = 3
+HOLD_WAIT_WINDOW_MIN = 60
+
+
+def dispatch_hold_reason(agent, now=None, hold_path=None, job_log_path=None):
+    # Hold reason if dispatch to agent must be skipped, else None.
+    # Explicit operator holds win; otherwise auto-hold after repeated waits.
+    from datetime import timedelta
+    now = now or datetime.now(timezone.utc)
+    try:
+        with open(hold_path or DISPATCH_HOLD_FILE) as f:
+            holds = json.load(f)
+    except (OSError, ValueError):
+        holds = {}
+    entry = holds.get(agent) if isinstance(holds, dict) else None
+    if isinstance(entry, dict):
+        until = entry.get("until")
+        if until:
+            try:
+                exp = datetime.fromisoformat(until)
+                if exp.tzinfo is None:
+                    exp = exp.replace(tzinfo=timezone.utc)
+            except ValueError:
+                exp = None
+            if exp is not None and exp <= now:
+                entry = None
+        if entry is not None:
+            return "explicit hold (%s)" % entry.get("reason", "operator")
+    try:
+        cutoff = now - timedelta(minutes=HOLD_WAIT_WINDOW_MIN)
+        n = 0
+        with open(job_log_path or JOB_LOG) as f:
+            for line in f:
+                try:
+                    r = json.loads(line)
+                except ValueError:
+                    continue
+                if r.get("type") != "job_dispatch_agent_input_wait":
+                    continue
+                if r.get("agent") != agent:
+                    continue
+                try:
+                    ts = datetime.fromisoformat(r.get("ts", ""))
+                except ValueError:
+                    continue
+                if ts.tzinfo is None:
+                    ts = ts.replace(tzinfo=timezone.utc)
+                if ts >= cutoff:
+                    n += 1
+        if n >= HOLD_WAIT_THRESHOLD:
+            return "auto-hold (%d input-waits in last %dm)" % (n, HOLD_WAIT_WINDOW_MIN)
+    except OSError:
+        pass
+    return None
+
+
 def log_event(event_type, data):
     """Append event to job-log.jsonl"""
     entry = {
@@ -395,6 +495,13 @@ def main():
     # Load job
     job = load_job(job_name)
 
+    # Backpressure: skip frozen agents before arming follow-ups or sending.
+    _hold = dispatch_hold_reason(job.get("agent"))
+    if _hold:
+        print("Held: job %s for %s skipped (%s)." % (job_name, job.get("agent"), _hold), file=sys.stderr)
+        log_event("job_dispatch_held", {"job_name": job_name, "agent": job.get("agent"), "reason": _hold})
+        sys.exit(0)
+
     # Generate job_id
     job_id = f"{job_name}-{datetime.now(timezone.utc).strftime('%Y%m%d-%H%M%S')}-{uuid.uuid4().hex[:8]}"
 
@@ -461,19 +568,33 @@ def main():
         # Check if reuse_key exists in job-sidechats.json and thread is still alive
         sc_state = load_sidechat_state()
         reused_uuid = None
+        rotated_from = None
         if reuse_key and reuse_key in sc_state:
             val = sc_state[reuse_key]
             cand_uuid = val.get("thread_uuid") if isinstance(val, dict) else val
             if cand_uuid:
-                try:
-                    import muse_hybrid
-                    threads, err = muse_hybrid.get_threads(agent)
-                    if not err and threads:
-                        thread_ids = [t.get("session_id") for t in threads]
-                        if cand_uuid in thread_ids:
-                            reused_uuid = cand_uuid
-                except Exception:
-                    pass
+                rotate, reason = should_rotate_sidechat(val, sc_name)
+                if rotate:
+                    print(f"Rotating sidechat '{reuse_key}': {reason}")
+                    log_event("job_sidechat_rotate", {
+                        "job_name": job_name, "job_id": job_id,
+                        "reuse_key": reuse_key, "old_thread": cand_uuid,
+                        "reason": reason,
+                    })
+                    rotated_from = cand_uuid
+                else:
+                    try:
+                        import muse_hybrid
+                        threads, err = muse_hybrid.get_threads(agent)
+                        if not err and threads:
+                            thread_ids = [t.get("session_id") for t in threads]
+                            if cand_uuid in thread_ids:
+                                reused_uuid = cand_uuid
+                    except Exception:
+                        pass
+                    if reused_uuid and isinstance(val, dict):
+                        val["dispatch_count"] = val.get("dispatch_count", 0) + 1
+                        save_sidechat_state(sc_state)
 
         if reused_uuid:
             target = reused_uuid
@@ -487,13 +608,19 @@ def main():
                     new_uuid = res.get("session_id")
                     key_to_save = reuse_key or sc_name
                     is_persistent = bool(reuse_key)
-                    sc_state[key_to_save] = {
+                    new_record = {
                         "thread_uuid": new_uuid,
                         "agent": agent,
                         "title": channel_title,
                         "type": "persistent" if is_persistent else "ephemeral",
-                        "created_at": datetime.now(timezone.utc).isoformat()
+                        "created_at": datetime.now(timezone.utc).isoformat(),
+                        "dispatch_count": 1,
                     }
+                    if rotated_from:
+                        new_record["rotated_from"] = rotated_from
+                        new_record["rotated_at"] = datetime.now(
+                            timezone.utc).isoformat()
+                    sc_state[key_to_save] = new_record
                     save_sidechat_state(sc_state)
                     target = new_uuid
                     print(f"Spawned new sidechat channel '{channel_title}' ({new_uuid}) for {agent}")