diff --git a/cron/scheduler.py b/cron/scheduler.py
index 44689d8d0..b2b18824e 100644
--- a/cron/scheduler.py
+++ b/cron/scheduler.py
@@ -1244,10 +1244,11 @@ def _run_via_claude_subprocess(
         "TZ": os.environ.get("TZ", "Asia/Seoul"),
     }
 
-    logger.info("Job '%s': running via claude -p subprocess (subscription)", job_id)
+    _model = os.environ.get("HERMES_CLAUDE_SUB_MODEL", "claude-opus-4-8").strip() or "claude-opus-4-8"
+    logger.info("Job '%s': running via claude -p subprocess (subscription) model=%s", job_id, _model)
     try:
         result = subprocess.run(
-            [claude_bin, "-p", prompt],
+            [claude_bin, "-p", "--model", _model, prompt],
             capture_output=True,
             text=True,
             env=env,
diff --git a/gateway/run.py b/gateway/run.py
index f0c7190a0..23d979056 100644
--- a/gateway/run.py
+++ b/gateway/run.py
@@ -17829,23 +17829,29 @@ def _interim_assistant_cb(text: str, *, already_streamed: bool = False) -> None:
             ):
                 import subprocess as _subprocess
                 import shutil as _shutil
+                import uuid as _uuid
                 _claude_bin = _shutil.which("claude")
                 if _claude_bin:
-                    _hist_lines = []
-                    for _h in (history or []):
-                        _role = str(_h.get("role") or "")
-                        _content = _h.get("content") or ""
-                        if isinstance(_content, list):
-                            _content = " ".join(
-                                c.get("text", "") if isinstance(c, dict) else str(c)
-                                for c in _content
-                            ).strip()
-                        if _role and _content:
-                            _hist_lines.append(f"{_role.capitalize()}: {_content}")
-                    _full_prompt = (
-                        "[Previous conversation:\n" + "\n".join(_hist_lines) + "\n]\n\nUser: " + message
-                        if _hist_lines else message
-                    )
+                    # --- session-id mapping: gateway session_key → claude session uuid ---
+                    _session_map_path = Path.home() / ".claude-hermes" / "gateway_session_map.json"
+                    _session_map: dict = {}
+                    try:
+                        if _session_map_path.exists():
+                            _session_map = json.loads(_session_map_path.read_text())
+                    except Exception:
+                        _session_map = {}
+                    _existing_session_id = _session_map.get(session_key or "")
+                    _is_new_session = not _existing_session_id
+                    if _is_new_session:
+                        _claude_session_id = str(_uuid.uuid4())
+                        _session_map[session_key or ""] = _claude_session_id
+                        try:
+                            _session_map_path.write_text(json.dumps(_session_map, ensure_ascii=False))
+                        except Exception:
+                            pass
+                    else:
+                        _claude_session_id = _existing_session_id
+                    # --- build subprocess command ---
                     _sub_env = {
                         "PATH": os.environ.get("PATH", ""),
                         "HOME": os.environ.get("HOME", ""),
@@ -17858,9 +17864,12 @@ def _interim_assistant_cb(text: str, *, already_streamed: bool = False) -> None:
                     }
                     _sub_timeout = float(os.getenv("HERMES_CRON_TIMEOUT", "600") or "600")
                     _sub_model = str(turn_route.get("model") or "").strip()
-                    _sub_cmd = [_claude_bin, "-p", _full_prompt]
+                    # new session → --session-id to create; existing → --resume to continue
+                    _session_flag = "--session-id" if _is_new_session else "--resume"
+                    _sub_cmd = [_claude_bin, "-p", _session_flag, _claude_session_id]
                     if _sub_model and _sub_model.startswith("claude"):
-                        _sub_cmd = [_claude_bin, "-p", "--model", _sub_model, _full_prompt]
+                        _sub_cmd += ["--model", _sub_model]
+                    _sub_cmd.append(message)
                     try:
                         _sub_result = _subprocess.run(
                             _sub_cmd,