diff --git a/contributors/emails/ait0u5hi@users.noreply.github.com b/contributors/emails/ait0u5hi@users.noreply.github.com
new file mode 100644
index 0000000000000..03959dcdfdfd7
--- /dev/null
+++ b/contributors/emails/ait0u5hi@users.noreply.github.com
@@ -0,0 +1,2 @@
+Ait0u5hi
+# cron process-group reaping fix
diff --git a/cron/scheduler.py b/cron/scheduler.py
index c5957070021e7..6087edcc66b77 100644
--- a/cron/scheduler.py
+++ b/cron/scheduler.py
@@ -17,6 +17,7 @@
 import os
 import re
 import shutil
+import signal
 import subprocess
 import sys
 import threading
@@ -2204,19 +2205,45 @@ def _run_job_script(script_path: str) -> tuple[bool, str]:
                 "encoding": "utf-8",
                 "errors": "replace",
             }
+        else:
+            # Own session/process-group so a timeout can reap the WHOLE tree.
+            # subprocess.run's timeout SIGKILLs only the direct child, orphaning
+            # backgrounded grandchildren; a child wedged in D-state then hangs
+            # the reap, leaking stuck processes (the 2026-07-15 tegrastats
+            # incident class). A new session lets us kill the process group.
+            popen_kwargs["start_new_session"] = True
         env = _sanitize_subprocess_env(os.environ.copy())
         env.update(env_overlay)
-        result = subprocess.run(
+        proc = subprocess.Popen(
             argv,
-            capture_output=True,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
             text=True,
-            timeout=script_timeout,
             cwd=str(path.parent),
             env=env,
             **popen_kwargs,
         )
-        stdout = (result.stdout or "").strip()
-        stderr = (result.stderr or "").strip()
+        try:
+            out, err = proc.communicate(timeout=script_timeout)
+        except subprocess.TimeoutExpired:
+            # Kill the entire process group (backgrounded grandchildren too),
+            # not just the direct child. Best-effort on a wedged/D-state tree:
+            # SIGKILL the group, then reap without blocking forever.
+            if sys.platform != "win32":
+                try:
+                    os.killpg(os.getpgid(proc.pid), getattr(signal, "SIGKILL", signal.SIGTERM))  # windows-footgun: ok
+                except (ProcessLookupError, PermissionError, OSError):
+                    pass
+            else:
+                proc.kill()
+            try:
+                proc.communicate(timeout=5)
+            except Exception:
+                pass
+            raise
+        returncode = proc.returncode
+        stdout = (out or "").strip()
+        stderr = (err or "").strip()
 
         # Redact secrets from both stdout and stderr before any return path.
         try:
@@ -2228,8 +2255,8 @@ def _run_job_script(script_path: str) -> tuple[bool, str]:
             stdout = "[REDACTED - redaction failed]"
             stderr = "[REDACTED - redaction failed]"
 
-        if result.returncode != 0:
-            parts = [f"Script exited with code {result.returncode}"]
+        if returncode != 0:
+            parts = [f"Script exited with code {returncode}"]
             if stderr:
                 parts.append(f"stderr:\n{stderr}")
             if stdout:
diff --git a/tests/cron/test_scheduler_process_group_reaping.py b/tests/cron/test_scheduler_process_group_reaping.py
new file mode 100644
index 0000000000000..93c8b1fe25cde
--- /dev/null
+++ b/tests/cron/test_scheduler_process_group_reaping.py
@@ -0,0 +1,51 @@
+"""Cron script jobs must reap the whole process GROUP on timeout.
+
+`subprocess.run(timeout=...)` SIGKILLs only the direct child, so a backgrounded
+grandchild is orphaned and a child wedged in D-state hangs the reap. `_run_job_script`
+now runs the script in its own session/process group and, on TimeoutExpired,
+`os.killpg`s the whole group. This is the regression test for that behavior.
+"""
+import subprocess
+import sys
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+import cron.scheduler as sched
+from cron.scheduler import _run_job_script
+
+
+@pytest.mark.skipif(
+    sys.platform == "win32",
+    reason="process-group reaping is POSIX-only; Windows uses proc.kill()",
+)
+def test_timeout_kills_whole_process_group(tmp_path, monkeypatch):
+    # Point the scripts dir at a temp HERMES_HOME and drop a valid .sh in it.
+    monkeypatch.setattr(sched, "_get_hermes_home", lambda: tmp_path)
+    scripts = tmp_path / "scripts"
+    scripts.mkdir()
+    (scripts / "reap.sh").write_text("#!/bin/bash\nsleep 300 &\nsleep 300\n")
+
+    fake = MagicMock()
+    fake.pid = 4242
+    fake.returncode = -9
+    # First communicate() (with the job timeout) wedges; the post-kill reap returns.
+    fake.communicate.side_effect = [
+        subprocess.TimeoutExpired(cmd="reap.sh", timeout=1),
+        ("", ""),
+    ]
+
+    with patch("cron.scheduler.subprocess.Popen", return_value=fake) as popen, \
+            patch("cron.scheduler.os.getpgid", return_value=4242) as getpgid, \
+            patch("cron.scheduler.os.killpg") as killpg:
+        ok, msg = _run_job_script("reap.sh")
+
+    # Child is launched in its own session so the group is killable...
+    assert popen.call_args.kwargs.get("start_new_session") is True
+    # ...and the WHOLE group is killed on timeout, not just the direct child.
+    getpgid.assert_called_once_with(4242)
+    killpg.assert_called_once()
+    assert killpg.call_args.args[0] == 4242
+    # The timeout is still reported back to the caller.
+    assert ok is False
+    assert "timed out" in msg.lower()