test(tui_gateway): make the tui gateway suite order-independent

Cross-file leaks made tests/tui_gateway + tests/test_tui_gateway_server.py
fail when run in one process (issue #57068):

- conftest: _hermetic_environment now re-pins hermes_state.DEFAULT_DB_PATH
  to the fake home so no test ever touches the developer's real state.db
- conftest: new autouse _reset_tui_gateway_server_state snapshots/restores
  _methods, cfg cache, db handle and _real_stdout, and tears down leftover
  sessions via the production _close_session_by_id(..., end_reason=
  "test_cleanup") boundary
- per-file fixtures scope patch.dict(sys.modules) to the import only
- drop reload()-based teardowns that duplicated atexit hooks

Conflict resolution vs main: kept main's mod._live_transports.clear() in
the test_protocol.py server fixture alongside the snapshot/restore logic.

Combined run now 792 passed / 0 failed.

Salvaged from PR #57066 by @lEWFkRAD. Fixes #57068.

pyproject.toml

@@ -367,6 +367,9 @@ markers = [
     "real_concurrent_gate: opt out of the autouse stub that disables _detect_concurrent_hermes_instances",
     "real_agent_prewarm: opt out of the autouse stub that disables the tui_gateway deferred agent pre-warm timer",
     "requires_wal: needs the runtime to actually enable SQLite WAL mode (skipped where Hermes falls back to journal_mode=DELETE)",
+    "no_isolate: opt out of per-file subprocess isolation (tests share mutable module-level state)",
+    "ssh: marks tests requiring a reachable SSH server (skipped in normal CI)",
+    "live_system_guard_bypass: opt out of the os.kill monkeypatch guard for tests that need real signal delivery",
 ]
 # integration tests take way too long to run in the normal CI environments
 addopts = "-m 'not integration'"

tests/conftest.py

@@ -1139,6 +1139,12 @@ def _live_system_guard(request, monkeypatch):
         "daemon-reload", "try-restart", "reload-or-restart",
     )
     _PROCESS_KILLERS = ("pkill", "killall", "taskkill", "skill", "fuser")
+    # Shell/launcher executables whose arguments are themselves commands —
+    # argv[0]-only scanning must not exempt what they wrap.
+    _WRAPPER_COMMANDS = (
+        "sh", "bash", "zsh", "dash", "env", "nohup", "setsid",
+        "timeout", "sudo", "xargs", "nice", "ionice", "stdbuf", "flock",
+    )
 
     def _cmd_to_string(cmd) -> str:
         if cmd is None:
@@ -1181,7 +1187,17 @@ def _live_system_guard(request, monkeypatch):
             tokens = cmd_str.split()
         if not tokens:
             return False
-        for tok in tokens:
+
+        # For argv-style calls only argv[0] is the executable; scanning every
+        # argument blocked innocent commands like ``cat /tmp/.../skill``
+        # ("skill" is in _PROCESS_KILLERS).  Wrapper executables still get
+        # full-token scanning so ``["bash", "-c", "pkill ..."]`` stays caught.
+        if isinstance(cmd, (list, tuple)):
+            head0 = tokens[0].rsplit("/", 1)[-1].rsplit("\\", 1)[-1]
+            killer_tokens = tokens if head0 in _WRAPPER_COMMANDS else tokens[:1]
+        else:
+            killer_tokens = tokens
+        for tok in killer_tokens:
             head = tok.rsplit("/", 1)[-1].rsplit("\\", 1)[-1]
             if head in _PROCESS_KILLERS:
                 low = cmd_str.lower()

tests/test_live_system_guard.py

@@ -0,0 +1,39 @@
+"""Regression tests for the conftest live-system guard's argv handling.
+
+The guard must treat only argv[0] of a list/tuple command as the executable
+(arguments are data: a file named ``skill`` is not the ``skill`` binary),
+while still scanning every token of wrapper invocations like ``bash -c``.
+All blocked-case commands use patterns that match no real process, so a
+guard regression cannot kill anything.
+"""
+
+import subprocess
+
+import pytest
+
+
+def test_argv_arguments_are_not_treated_as_executables(tmp_path):
+    """A file argument whose basename is a killer name must not trip the
+    guard (the path contains "hermes" via the pytest tmp root)."""
+    target = tmp_path / "skill"
+    target.write_text("just a filename\n")
+    result = subprocess.run(["cat", str(target)], capture_output=True, text=True)
+    assert result.returncode == 0
+    assert "just a filename" in result.stdout
+
+
+def test_direct_killer_argv_is_still_blocked():
+    with pytest.raises(RuntimeError, match="live-system guard"):
+        subprocess.run(["pkill", "-f", "hermes-guard-regression-nomatch"])
+
+
+def test_wrapped_killer_command_is_still_blocked():
+    """argv[0]-only scanning must not exempt commands hidden behind a
+    shell wrapper."""
+    with pytest.raises(RuntimeError, match="live-system guard"):
+        subprocess.run(["bash", "-c", "pkill -f hermes-guard-regression-nomatch"])
+
+
+def test_env_wrapped_killer_command_is_still_blocked():
+    with pytest.raises(RuntimeError, match="live-system guard"):
+        subprocess.run(["env", "GUARD_TEST=1", "pkill", "-f", "hermes-guard-regression-nomatch"])

tests/tui_gateway/test_protocol.py

@@ -88,7 +88,6 @@ def test_shared_fixture_cleanup_uses_full_session_teardown(server, monkeypatch):
 
     assert server._sessions == {}
     assert closed == {"worker": 1, "agent": 1, "lease": 1}
->>>>>>> theirs
 
 
 @pytest.fixture()