Merge upstream/main and address Copilot review feedback

Merge resolved conflicts in web/src/{i18n/{en,zh,types}.ts,lib/api.ts}
by keeping both this branch's `profiles` additions and upstream's new
`models` page additions.

Copilot review feedback:
- Implement POST /api/profiles/{name}/open-terminal endpoint (already
  present); align Windows branch to `cmd.exe /c start "" <cmd>` so it
  matches the new test and spawns a fresh window instead of /k reusing
  the parent console.
- Move backslash escaping out of the macOS AppleScript f-string
  expression (Python <3.12 disallows backslashes inside f-string
  expression parts).
- Patch `_get_wrapper_dir` via monkeypatch in
  test_profiles_create_creates_wrapper_alias_when_safe so the test no
  longer writes to the real `~/.local/bin`.
- Extend test_dashboard_browser_safe_imports to scan `.ts` files in
  addition to `.tsx`.
- Switch upstream's new ModelsPage.tsx away from the `@nous-research/ui`
  root barrel onto per-component subpaths to satisfy the stricter scan.
- Fix NouiTypography `leading-1.4` -> `leading-[1.4]` so Tailwind
  actually emits the line-height for the `sm` variant.
- Guard ProfilesPage.openSoulEditor against out-of-order responses by
  tracking the latest requested profile via a ref.
- Replace ProfilesPage's hand-rolled setup command with a fetch to
  `/api/profiles/{name}/setup-command` so the copied command always
  matches what the backend would actually run (handles wrapper-alias
  collisions and reserved names correctly).
- Wire SOUL.md textarea label `htmlFor` -> textarea `id` so screen
  readers and clicking the label work as expected.

tests/hermes_cli/test_api_key_providers.py

@@ -1097,3 +1097,63 @@ class TestHuggingFaceModels:
         from hermes_cli.models import _PROVIDER_LABELS
         assert "huggingface" in _PROVIDER_LABELS
         assert _PROVIDER_LABELS["huggingface"] == "Hugging Face"
+
+
+# =============================================================================
+# MiniMax OAuth provider tests (added by feat/minimax-oauth-provider)
+# =============================================================================
+
+class TestMinimaxOAuthProvider:
+    """Tests for the minimax-oauth OAuth provider."""
+
+    def test_minimax_oauth_in_provider_registry(self):
+        assert "minimax-oauth" in PROVIDER_REGISTRY
+        pconfig = PROVIDER_REGISTRY["minimax-oauth"]
+        assert pconfig.auth_type == "oauth_minimax"
+        assert pconfig.id == "minimax-oauth"
+
+    def test_minimax_oauth_has_correct_endpoints(self):
+        from hermes_cli.auth import (
+            MINIMAX_OAUTH_GLOBAL_BASE,
+            MINIMAX_OAUTH_GLOBAL_INFERENCE,
+            MINIMAX_OAUTH_CN_BASE,
+            MINIMAX_OAUTH_CN_INFERENCE,
+        )
+        pconfig = PROVIDER_REGISTRY["minimax-oauth"]
+        assert pconfig.portal_base_url == MINIMAX_OAUTH_GLOBAL_BASE
+        assert pconfig.inference_base_url == MINIMAX_OAUTH_GLOBAL_INFERENCE
+        assert pconfig.extra["cn_portal_base_url"] == MINIMAX_OAUTH_CN_BASE
+        assert pconfig.extra["cn_inference_base_url"] == MINIMAX_OAUTH_CN_INFERENCE
+
+    def test_minimax_oauth_alias_resolves_portal(self):
+        result = resolve_provider("minimax-portal")
+        assert result == "minimax-oauth"
+
+    def test_minimax_oauth_alias_resolves_global(self):
+        result = resolve_provider("minimax-global")
+        assert result == "minimax-oauth"
+
+    def test_minimax_oauth_alias_resolves_underscore(self):
+        result = resolve_provider("minimax_oauth")
+        assert result == "minimax-oauth"
+
+    def test_minimax_oauth_listed_in_canonical_providers(self):
+        from hermes_cli.models import CANONICAL_PROVIDERS
+        slugs = [p.slug for p in CANONICAL_PROVIDERS]
+        assert "minimax-oauth" in slugs
+
+    def test_minimax_oauth_models_alias_in_models_py(self):
+        from hermes_cli.models import _PROVIDER_ALIASES
+        assert _PROVIDER_ALIASES.get("minimax-portal") == "minimax-oauth"
+        assert _PROVIDER_ALIASES.get("minimax-global") == "minimax-oauth"
+        assert _PROVIDER_ALIASES.get("minimax_oauth") == "minimax-oauth"
+
+    def test_minimax_oauth_has_models(self):
+        from hermes_cli.models import _PROVIDER_MODELS
+        models = _PROVIDER_MODELS.get("minimax-oauth", [])
+        assert len(models) >= 1
+
+    def test_minimax_oauth_aux_model_registered(self):
+        from agent.auxiliary_client import _API_KEY_PROVIDER_AUX_MODELS
+        assert "minimax-oauth" in _API_KEY_PROVIDER_AUX_MODELS
+        assert _API_KEY_PROVIDER_AUX_MODELS["minimax-oauth"]  # non-empty

tests/hermes_cli/test_auth_commands.py

@@ -1446,23 +1446,36 @@ def test_seed_custom_pool_respects_config_suppression(tmp_path, monkeypatch):
 def test_credential_sources_registry_has_expected_steps():
     """Sanity check — the registry contains the expected RemovalSteps.
 
-    Guards against accidentally dropping a step during future refactors.
-    If you add a new credential source, add it to the expected set below.
+    Adding a new credential source is routine, so this is a structural
+    invariant check (every step has a description, every step is unique,
+    core steps are present) rather than a frozen snapshot. Frozen
+    snapshots of catalog-like data violate the AGENTS.md "don't write
+    change-detector tests" rule — they break every time someone adds a
+    provider.
     """
     from agent.credential_sources import _REGISTRY
 
-    descriptions = {step.description for step in _REGISTRY}
-    expected = {
+    descriptions = [step.description for step in _REGISTRY]
+    # No empty descriptions, no duplicates.
+    assert all(d for d in descriptions), "Every removal step must have a description"
+    assert len(descriptions) == len(set(descriptions)), (
+        f"Registry has duplicate step descriptions: {descriptions}"
+    )
+    # Core steps must be present — these are the ones the rest of the code
+    # assumes exist. When deliberately dropping one, update this list.
+    required = {
         "gh auth token / COPILOT_GITHUB_TOKEN / GH_TOKEN",
         "Any env-seeded credential (XAI_API_KEY, DEEPSEEK_API_KEY, etc.)",
         "~/.claude/.credentials.json",
         "~/.hermes/.anthropic_oauth.json",
         "auth.json providers.nous",
         "auth.json providers.openai-codex + ~/.codex/auth.json",
+        "auth.json providers.minimax-oauth",
         "~/.qwen/oauth_creds.json",
         "Custom provider config.yaml api_key field",
     }
-    assert descriptions == expected, f"Registry mismatch. Got: {descriptions}"
+    missing = required - set(descriptions)
+    assert not missing, f"Registry missing required steps: {missing}"
 
 
 def test_credential_sources_find_step_returns_none_for_manual():

tests/hermes_cli/test_claw.py

@@ -526,6 +526,11 @@ class TestCmdMigrate:
 class TestCmdCleanup:
     """Test the cleanup command handler."""
 
+    @pytest.fixture(autouse=True)
+    def _mock_openclaw_running(self):
+        with patch.object(claw_mod, "_detect_openclaw_processes", return_value=[]):
+            yield
+
     def test_no_dirs_found(self, tmp_path, capsys):
         args = Namespace(source=None, dry_run=False, yes=False)
         with patch.object(claw_mod, "_find_openclaw_dirs", return_value=[]):

tests/hermes_cli/test_config_env_expansion.py

@@ -72,7 +72,10 @@ class TestLoadConfigExpansion:
 
         monkeypatch.setenv("GOOGLE_API_KEY", "gsk-test-key")
         monkeypatch.setenv("TELEGRAM_BOT_TOKEN", "1234567:ABC-token")
-        monkeypatch.setattr("hermes_cli.config.get_config_path", lambda: config_file)
+        # Patch the imported function's own globals. Other tests may reload
+        # hermes_cli.config, making string-target monkeypatches hit a different
+        # module object than this collection-time imported load_config().
+        monkeypatch.setitem(load_config.__globals__, "get_config_path", lambda: config_file)
 
         config = load_config()
 
@@ -86,7 +89,7 @@ class TestLoadConfigExpansion:
         config_file.write_text(config_yaml)
 
         monkeypatch.delenv("NOT_SET_XYZ_123", raising=False)
-        monkeypatch.setattr("hermes_cli.config.get_config_path", lambda: config_file)
+        monkeypatch.setitem(load_config.__globals__, "get_config_path", lambda: config_file)
 
         config = load_config()
 

tests/hermes_cli/test_container_aware_cli.py

@@ -105,7 +105,7 @@ def test_get_container_exec_info_defaults():
         )
 
         with patch("hermes_constants.is_container", return_value=False), \
-             patch("hermes_cli.config.get_hermes_home", return_value=hermes_home), \
+             patch.dict(get_container_exec_info.__globals__, {"get_hermes_home": lambda: hermes_home}), \
              patch.dict(os.environ, {}, clear=False):
             os.environ.pop("HERMES_DEV", None)
             info = get_container_exec_info()

tests/hermes_cli/test_dashboard_browser_safe_imports.py

@@ -7,9 +7,10 @@ WEB_SRC = Path(__file__).resolve().parents[2] / "web" / "src"
 
 def test_dashboard_does_not_import_nous_ui_root_barrel():
     offenders = []
-    for path in WEB_SRC.rglob("*.tsx"):
-        content = path.read_text(encoding="utf-8")
-        if 'from "@nous-research/ui"' in content or "from '@nous-research/ui'" in content:
-            offenders.append(str(path.relative_to(WEB_SRC)))
+    for ext in ("*.tsx", "*.ts"):
+        for path in WEB_SRC.rglob(ext):
+            content = path.read_text(encoding="utf-8")
+            if 'from "@nous-research/ui"' in content or "from '@nous-research/ui'" in content:
+                offenders.append(str(path.relative_to(WEB_SRC)))
 
     assert offenders == []

tests/hermes_cli/test_dashboard_lifecycle_flags.py

@@ -0,0 +1,181 @@
+"""Tests for ``hermes dashboard --stop`` / ``--status`` flags.
+
+These flags share the detection + kill path with the post-``hermes update``
+cleanup, so the heavy coverage of SIGTERM / SIGKILL / Windows taskkill lives
+in ``test_update_stale_dashboard.py``.  This file just verifies the flag
+dispatch: argparse wiring, no-op when nothing is running, and correct
+exit codes.
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+from unittest.mock import patch, MagicMock
+
+import pytest
+
+from hermes_cli.main import cmd_dashboard, _report_dashboard_status
+
+
+def _ns(**kw):
+    """Build an argparse.Namespace with dashboard defaults plus overrides."""
+    defaults = dict(
+        port=9119, host="127.0.0.1", no_open=False, insecure=False,
+        tui=False, stop=False, status=False,
+    )
+    defaults.update(kw)
+    return argparse.Namespace(**defaults)
+
+
+class TestDashboardStatus:
+    def test_status_no_processes(self, capsys):
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(status=True))
+        assert exc.value.code == 0
+        out = capsys.readouterr().out
+        assert "No hermes dashboard processes running" in out
+
+    def test_status_with_processes(self, capsys):
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345, 12346]), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(status=True))
+        # Status is informational — always exits 0.
+        assert exc.value.code == 0
+        out = capsys.readouterr().out
+        assert "2 hermes dashboard process(es) running" in out
+        assert "PID 12345" in out
+        assert "PID 12346" in out
+
+    def test_status_does_not_try_to_import_fastapi(self):
+        """`--status` must not require dashboard runtime deps — it's a
+        process-table scan only.  We prove this by making fastapi import
+        fail and confirming --status still succeeds."""
+        orig_import = __import__
+        def fake_import(name, *a, **kw):
+            if name == "fastapi":
+                raise ImportError("fastapi missing")
+            return orig_import(name, *a, **kw)
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             patch("builtins.__import__", side_effect=fake_import), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(status=True))
+        assert exc.value.code == 0
+
+
+class TestDashboardStop:
+    def test_stop_when_nothing_running(self, capsys):
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(stop=True))
+        assert exc.value.code == 0
+        out = capsys.readouterr().out
+        assert "No hermes dashboard processes running" in out
+
+    def test_stop_kills_and_exits_zero_when_all_killed(self, capsys):
+        """After the kill, if the second scan returns empty we exit 0."""
+        # First scan: finds two processes.  Second (verification) scan: empty.
+        scans = iter([[12345, 12346], []])
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   side_effect=lambda: next(scans)), \
+             patch("hermes_cli.main._kill_stale_dashboard_processes") as mock_kill, \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(stop=True))
+        mock_kill.assert_called_once()
+        # --stop should pass a reason so the output doesn't say "running
+        # backend no longer matches the updated frontend" (that wording is
+        # for the post-`hermes update` path).
+        kwargs = mock_kill.call_args.kwargs
+        assert "reason" in kwargs
+        assert "stop" in kwargs["reason"].lower()
+        assert exc.value.code == 0
+
+    def test_stop_exits_nonzero_if_kill_leaves_survivors(self):
+        """If the second scan still finds PIDs, we exit 1 so scripts can
+        detect that the stop didn't succeed (e.g. permission denied)."""
+        scans = iter([[12345], [12345]])  # both scans find the same PID
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   side_effect=lambda: next(scans)), \
+             patch("hermes_cli.main._kill_stale_dashboard_processes"), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(stop=True))
+        assert exc.value.code == 1
+
+    def test_stop_does_not_try_to_import_fastapi(self):
+        """Like --status, --stop must work without dashboard runtime deps."""
+        orig_import = __import__
+        def fake_import(name, *a, **kw):
+            if name == "fastapi":
+                raise ImportError("fastapi missing")
+            return orig_import(name, *a, **kw)
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             patch("builtins.__import__", side_effect=fake_import), \
+             pytest.raises(SystemExit) as exc:
+            cmd_dashboard(_ns(stop=True))
+        assert exc.value.code == 0
+
+
+class TestLifecycleFlagsTakePrecedence:
+    """If both --stop and --status are set, --status wins (it's listed
+    first in cmd_dashboard).  Neither is allowed to fall through to the
+    server-start path, which is the critical safety property — a user
+    who typed ``hermes dashboard --stop`` must not end up ALSO starting
+    a new server."""
+
+    def test_status_wins_over_stop(self, capsys):
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             patch("hermes_cli.main._kill_stale_dashboard_processes") as mock_kill, \
+             pytest.raises(SystemExit):
+            cmd_dashboard(_ns(status=True, stop=True))
+        # Kill path must NOT run when --status is also set.
+        mock_kill.assert_not_called()
+
+    def test_stop_does_not_fall_through_to_server_start(self):
+        """Covers the worst-case regression: if --stop ever stopped exiting
+        early, the user would start the dashboard they just asked to stop."""
+        called = {"start": False}
+        def fake_start_server(**kw):
+            called["start"] = True
+
+        # Provide a fake web_server module so the import doesn't matter.
+        fake_ws = MagicMock()
+        fake_ws.start_server = fake_start_server
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             patch.dict(sys.modules, {"hermes_cli.web_server": fake_ws}), \
+             pytest.raises(SystemExit):
+            cmd_dashboard(_ns(stop=True))
+        assert called["start"] is False
+
+
+class TestArgparseWiring:
+    """Confirm the flags are exposed via the real argparse tree so
+    ``hermes dashboard --stop`` / ``--status`` actually parse."""
+
+    def test_flags_are_registered(self):
+        from hermes_cli.main import main as _cli_main  # noqa: F401
+        # Rebuild the argparse tree by re-running the section of main()
+        # that builds it.  Cheapest way: introspect via --help on the
+        # already-built parser would require refactoring; instead we
+        # parse the flags directly via a minimal replay.
+        import importlib
+        mod = importlib.import_module("hermes_cli.main")
+        # Find the dashboard_parser instance by running build logic would
+        # be too invasive.  Instead parse args as if via the CLI by
+        # intercepting parse_args.  This is overkill for a smoke test —
+        # we just want to know the flags don't KeyError.
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[]), \
+             pytest.raises(SystemExit) as exc:
+            mod.cmd_dashboard(_ns(status=True))
+        assert exc.value.code == 0

tests/hermes_cli/test_doctor.py

@@ -161,6 +161,38 @@ def test_check_gateway_service_linger_skips_when_service_not_installed(monkeypat
     assert issues == []
 
 
+def test_doctor_reports_vercel_backend_diagnostics(monkeypatch, tmp_path):
+    monkeypatch.setenv("TERMINAL_ENV", "vercel_sandbox")
+    monkeypatch.setenv("TERMINAL_VERCEL_RUNTIME", "python3.13")
+    monkeypatch.setenv("TERMINAL_CONTAINER_DISK", "2048")
+    monkeypatch.setenv("VERCEL_TOKEN", "super-secret-value")
+    monkeypatch.delenv("VERCEL_PROJECT_ID", raising=False)
+    monkeypatch.setenv("VERCEL_TEAM_ID", "team")
+    monkeypatch.setattr(doctor_mod.importlib.util, "find_spec", lambda name: object() if name == "vercel" else None)
+
+    fake_model_tools = types.SimpleNamespace(
+        check_tool_availability=lambda *a, **kw: ([], []),
+        TOOLSET_REQUIREMENTS={},
+    )
+    monkeypatch.setitem(sys.modules, "model_tools", fake_model_tools)
+
+    buf = io.StringIO()
+    with contextlib.redirect_stdout(buf):
+        doctor_mod.run_doctor(Namespace(fix=False))
+
+    out = buf.getvalue()
+    assert "Vercel runtime" in out
+    assert "python3.13" in out
+    assert "Vercel custom disk unsupported" in out
+    assert "Vercel auth incomplete" in out
+    assert "VERCEL_PROJECT_ID" in out
+    assert "Vercel auth mode: incomplete access token" in out
+    assert "Vercel auth present env: VERCEL_TOKEN, VERCEL_TEAM_ID" in out
+    assert "Vercel auth missing env: VERCEL_PROJECT_ID" in out
+    assert "super-secret-value" not in out
+    assert "snapshot filesystem only" in out
+
+
 # ── Memory provider section (doctor should only check the *active* provider) ──
 
 

tests/hermes_cli/test_gateway_service.py

@@ -14,6 +14,26 @@ from gateway.restart import (
 )
 
 
+class TestUserSystemdPrivateSocketPreflight:
+    def test_preflight_accepts_private_socket_without_dbus_bus(self, monkeypatch):
+        monkeypatch.setattr(gateway_cli, "_ensure_user_systemd_env", lambda: None)
+        monkeypatch.setattr(gateway_cli, "_user_dbus_socket_path", lambda: Path("/tmp/missing-bus"))
+        monkeypatch.setattr(gateway_cli, "_user_systemd_private_socket_path", lambda: Path("/tmp/private-socket"))
+        monkeypatch.setattr(Path, "exists", lambda self: str(self) == "/tmp/private-socket")
+
+        gateway_cli._preflight_user_systemd(auto_enable_linger=False)
+
+    def test_wait_for_user_dbus_socket_accepts_private_socket(self, monkeypatch):
+        calls = []
+        monkeypatch.setattr(gateway_cli, "_ensure_user_systemd_env", lambda: calls.append("env"))
+        monkeypatch.setattr(gateway_cli, "_user_dbus_socket_path", lambda: Path("/tmp/missing-bus"))
+        monkeypatch.setattr(gateway_cli, "_user_systemd_private_socket_path", lambda: Path("/tmp/private-socket"))
+        monkeypatch.setattr(Path, "exists", lambda self: str(self) == "/tmp/private-socket")
+
+        assert gateway_cli._wait_for_user_dbus_socket(timeout=0.1) is True
+        assert calls == ["env"]
+
+
 class TestSystemdServiceRefresh:
     def test_systemd_install_repairs_outdated_unit_without_force(self, tmp_path, monkeypatch):
         unit_path = tmp_path / "hermes-gateway.service"
@@ -235,7 +255,8 @@ class TestLaunchdServiceRecovery:
         target = f"{domain}/{label}"
 
         def fake_run(cmd, check=False, **kwargs):
-            calls.append(cmd)
+            if cmd and cmd[0] == "launchctl":
+                calls.append(cmd)
             if cmd == ["launchctl", "kickstart", target] and calls.count(cmd) == 1:
                 raise gateway_cli.subprocess.CalledProcessError(3, cmd, stderr="Could not find service")
             return SimpleNamespace(returncode=0, stdout="", stderr="")
@@ -262,7 +283,8 @@ class TestLaunchdServiceRecovery:
         target = f"{domain}/{label}"
 
         def fake_run(cmd, check=False, **kwargs):
-            calls.append(cmd)
+            if cmd and cmd[0] == "launchctl":
+                calls.append(cmd)
             if cmd == ["launchctl", "kickstart", target] and calls.count(cmd) == 1:
                 raise gateway_cli.subprocess.CalledProcessError(113, cmd, stderr="Could not find service")
             return SimpleNamespace(returncode=0, stdout="", stderr="")
@@ -1105,6 +1127,10 @@ class TestPreflightUserSystemd:
             gateway_cli, "_user_dbus_socket_path",
             lambda: type("P", (), {"exists": lambda self: True})(),
         )
+        monkeypatch.setattr(
+            gateway_cli, "_user_systemd_private_socket_path",
+            lambda: type("P", (), {"exists": lambda self: False})(),
+        )
         # Should not raise, no subprocess calls needed.
         gateway_cli._preflight_user_systemd()
 
@@ -1114,6 +1140,10 @@ class TestPreflightUserSystemd:
             gateway_cli, "_user_dbus_socket_path",
             lambda: type("P", (), {"exists": lambda self: False})(),
         )
+        monkeypatch.setattr(
+            gateway_cli, "_user_systemd_private_socket_path",
+            lambda: type("P", (), {"exists": lambda self: False})(),
+        )
         monkeypatch.setattr(
             gateway_cli, "get_systemd_linger_status", lambda: (False, ""),
         )
@@ -1142,6 +1172,10 @@ class TestPreflightUserSystemd:
             gateway_cli, "_user_dbus_socket_path",
             lambda: type("P", (), {"exists": lambda self: False})(),
         )
+        monkeypatch.setattr(
+            gateway_cli, "_user_systemd_private_socket_path",
+            lambda: type("P", (), {"exists": lambda self: False})(),
+        )
         monkeypatch.setattr(
             gateway_cli, "get_systemd_linger_status",
             lambda: (None, "loginctl not found"),
@@ -1159,6 +1193,10 @@ class TestPreflightUserSystemd:
             gateway_cli, "_user_dbus_socket_path",
             lambda: type("P", (), {"exists": lambda self: False})(),
         )
+        monkeypatch.setattr(
+            gateway_cli, "_user_systemd_private_socket_path",
+            lambda: type("P", (), {"exists": lambda self: False})(),
+        )
         monkeypatch.setattr(
             gateway_cli, "get_systemd_linger_status", lambda: (True, ""),
         )
@@ -1177,6 +1215,10 @@ class TestPreflightUserSystemd:
             gateway_cli, "_user_dbus_socket_path",
             lambda: type("P", (), {"exists": lambda self: False})(),
         )
+        monkeypatch.setattr(
+            gateway_cli, "_user_systemd_private_socket_path",
+            lambda: type("P", (), {"exists": lambda self: False})(),
+        )
         monkeypatch.setattr(
             gateway_cli, "get_systemd_linger_status", lambda: (False, ""),
         )

tests/hermes_cli/test_ignore_user_config_flags.py

@@ -224,22 +224,21 @@ class TestArgparseFlagsRegistered:
         assert args.ignore_rules is True
 
     def test_main_py_registers_both_flags(self):
-        """E2E: the real hermes_cli/main.py parser accepts both flags.
+        """E2E: the real hermes parser accepts both flags."""
+        from hermes_cli._parser import build_top_level_parser
 
-        We invoke the real argparse tree builder from hermes_cli.main.
-        """
-        import hermes_cli.main as hm
+        parser, _subparsers, chat_parser = build_top_level_parser()
+
+        top_dests = {a.dest for a in parser._actions}
+        chat_dests = {a.dest for a in chat_parser._actions}
+        assert "ignore_user_config" in top_dests
+        assert "ignore_rules" in top_dests
+        assert "ignore_user_config" in chat_dests
+        assert "ignore_rules" in chat_dests
 
-        # hm has a helper that builds the argparse tree inside main().
-        # We can extract it by catching the SystemExit on --help.
-        # Simpler: just grep the source for the flag strings. Both approaches
-        # are brittle; we use a combined test.
-        import inspect
-        src = inspect.getsource(hm)
-        assert '"--ignore-user-config"' in src, \
-            "chat subparser must register --ignore-user-config"
-        assert '"--ignore-rules"' in src, \
-            "chat subparser must register --ignore-rules"
         # And the cmd_chat env-var wiring must be present
+        import inspect
+        import hermes_cli.main as hm
+        src = inspect.getsource(hm)
         assert "HERMES_IGNORE_USER_CONFIG" in src
         assert "HERMES_IGNORE_RULES" in src

tests/hermes_cli/test_mcp_reload_confirm_gate.py

@@ -0,0 +1,91 @@
+"""Tests for the approvals.mcp_reload_confirm config gate.
+
+When the user runs /reload-mcp, the MCP tool set is rebuilt which
+invalidates the provider prompt cache for the active session.  That's
+expensive on long-context / high-reasoning models.  The config gate
+adds a three-option confirmation (Approve Once / Always Approve /
+Cancel); "Always Approve" flips this key to false so subsequent reloads
+run silently.
+"""
+
+from __future__ import annotations
+
+from copy import deepcopy
+
+from hermes_cli.config import DEFAULT_CONFIG
+
+
+class TestMcpReloadConfirmDefault:
+    def test_default_config_has_the_key(self):
+        approvals = DEFAULT_CONFIG.get("approvals")
+        assert isinstance(approvals, dict)
+        assert "mcp_reload_confirm" in approvals
+
+    def test_default_is_true(self):
+        # New installs confirm by default — this is the safe behavior.
+        assert DEFAULT_CONFIG["approvals"]["mcp_reload_confirm"] is True
+
+    def test_shape_matches_other_approval_keys(self):
+        # Same flat dict level as `mode` / `timeout` / `cron_mode`.
+        approvals = DEFAULT_CONFIG["approvals"]
+        assert isinstance(approvals.get("mode"), str)
+        assert isinstance(approvals.get("timeout"), int)
+        assert isinstance(approvals.get("cron_mode"), str)
+        assert isinstance(approvals.get("mcp_reload_confirm"), bool)
+
+
+class TestUserConfigMerge:
+    """If a user has a pre-existing config without this key, load_config
+    should fill it in from DEFAULT_CONFIG (deep merge preserves keys the
+    user didn't override).
+    """
+
+    def test_existing_user_config_without_key_gets_default(self, tmp_path, monkeypatch):
+        import yaml
+
+        # Simulate a legacy user config without the new key.
+        home = tmp_path / ".hermes"
+        home.mkdir()
+        cfg_path = home / "config.yaml"
+        legacy = {
+            "approvals": {"mode": "manual", "timeout": 60, "cron_mode": "deny"},
+        }
+        cfg_path.write_text(yaml.safe_dump(legacy))
+
+        monkeypatch.setenv("HERMES_HOME", str(home))
+        # Force a fresh reimport of config.py so the HERMES_HOME is honored.
+        import importlib
+        import hermes_cli.config as cfg_mod
+        importlib.reload(cfg_mod)
+
+        cfg = cfg_mod.load_config()
+        assert cfg["approvals"]["mcp_reload_confirm"] is True
+
+    def test_existing_user_config_with_false_key_survives_merge(
+        self, tmp_path, monkeypatch,
+    ):
+        """A user who has clicked "Always Approve" (key=false) must keep
+        that setting across reloads — the default_true value must not win.
+        """
+        import yaml
+
+        home = tmp_path / ".hermes"
+        home.mkdir()
+        cfg_path = home / "config.yaml"
+        user_cfg = {
+            "approvals": {
+                "mode": "manual",
+                "timeout": 60,
+                "cron_mode": "deny",
+                "mcp_reload_confirm": False,
+            },
+        }
+        cfg_path.write_text(yaml.safe_dump(user_cfg))
+
+        monkeypatch.setenv("HERMES_HOME", str(home))
+        import importlib
+        import hermes_cli.config as cfg_mod
+        importlib.reload(cfg_mod)
+
+        cfg = cfg_mod.load_config()
+        assert cfg["approvals"]["mcp_reload_confirm"] is False

tests/hermes_cli/test_profiles.py

@@ -188,6 +188,23 @@ class TestCreateProfile:
         assert not (profile_dir / "gateway_state.json").exists()
         assert not (profile_dir / "processes.json").exists()
 
+    def test_clone_all_excludes_sibling_profiles_tree(self, profile_env):
+        """--clone-all from default ~/.hermes must not copy profiles/* (nested explosion)."""
+        tmp_path = profile_env
+        default_home = tmp_path / ".hermes"
+        profiles_root = default_home / "profiles"
+        profiles_root.mkdir(exist_ok=True)
+        (profiles_root / "other").mkdir(parents=True, exist_ok=True)
+        (profiles_root / "other" / "marker.txt").write_text("sibling data")
+
+        (default_home / "memories").mkdir(exist_ok=True)
+        (default_home / "memories" / "note.md").write_text("remember this")
+
+        profile_dir = create_profile("coder", clone_all=True, no_alias=True)
+
+        assert (profile_dir / "memories" / "note.md").read_text() == "remember this"
+        assert not (profile_dir / "profiles").exists()
+
     def test_clone_config_missing_files_skipped(self, profile_env):
         """Clone config gracefully skips files that don't exist in source."""
         profile_dir = create_profile("coder", clone_config=True, no_alias=True)

tests/hermes_cli/test_pty_bridge.py

@@ -96,10 +96,17 @@ class TestPtyBridgeIO:
 @skip_on_windows
 class TestPtyBridgeResize:
     def test_resize_updates_child_winsize(self):
-        # tput reads COLUMNS/LINES from the TTY ioctl (TIOCGWINSZ).
-        # Spawn a shell, resize, then ask tput for the dimensions.
+        # Query the TTY ioctl directly instead of using tput, which requires
+        # TERM and fails in GitHub Actions' non-interactive environment.
+        winsize_script = (
+            "import fcntl, struct, termios, time; "
+            "time.sleep(0.1); "
+            "rows, cols, *_ = struct.unpack('HHHH', "
+            "fcntl.ioctl(0, termios.TIOCGWINSZ, b'\\0' * 8)); "
+            "print(cols); print(rows)"
+        )
         bridge = PtyBridge.spawn(
-            ["/bin/sh", "-c", "sleep 0.1; tput cols; tput lines"],
+            [sys.executable, "-c", winsize_script],
             cols=80,
             rows=24,
         )

tests/hermes_cli/test_relaunch.py

@@ -0,0 +1,155 @@
+"""Tests for hermes_cli.relaunch — unified self-relaunch utility."""
+
+import sys
+
+import pytest
+
+from hermes_cli import relaunch as relaunch_mod
+
+
+class TestResolveHermesBin:
+    def test_prefers_absolute_argv0_when_executable(self, monkeypatch):
+        fake = "/nix/store/abc/bin/hermes"
+        monkeypatch.setattr(sys, "argv", [fake])
+        monkeypatch.setattr(relaunch_mod.os.path, "isfile", lambda p: p == fake)
+        monkeypatch.setattr(relaunch_mod.os, "access", lambda p, mode: p == fake)
+        assert relaunch_mod.resolve_hermes_bin() == fake
+
+    def test_resolves_relative_argv0(self, monkeypatch, tmp_path):
+        fake = tmp_path / "hermes"
+        fake.write_text("#!/bin/sh\n")
+        fake.chmod(0o755)
+        monkeypatch.setattr(sys, "argv", [str(fake.name)])
+        monkeypatch.chdir(tmp_path)
+        # Ensure we don't accidentally match a real 'hermes' on PATH
+        monkeypatch.setattr(relaunch_mod.shutil, "which", lambda _name: None)
+        assert relaunch_mod.resolve_hermes_bin() == str(fake)
+
+    def test_falls_back_to_path_which(self, monkeypatch):
+        monkeypatch.setattr(sys, "argv", ["-c"])  # not a real path
+        monkeypatch.setattr(
+            relaunch_mod.shutil, "which", lambda name: "/usr/bin/hermes" if name == "hermes" else None
+        )
+        assert relaunch_mod.resolve_hermes_bin() == "/usr/bin/hermes"
+
+    def test_returns_none_when_unresolvable(self, monkeypatch):
+        monkeypatch.setattr(sys, "argv", ["-c"])
+        monkeypatch.setattr(relaunch_mod.shutil, "which", lambda _name: None)
+        assert relaunch_mod.resolve_hermes_bin() is None
+
+
+class TestExtractInheritedFlags:
+    def test_extracts_tui_and_dev(self):
+        argv = ["--tui", "--dev", "chat"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["--tui", "--dev"]
+
+    def test_extracts_profile_with_value(self):
+        argv = ["--profile", "work", "chat"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["--profile", "work"]
+
+    def test_extracts_short_p_with_value(self):
+        argv = ["-p", "work"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["-p", "work"]
+
+    def test_extracts_equals_form(self):
+        argv = ["--profile=work", "--model=anthropic/claude-sonnet-4"]
+        assert relaunch_mod._extract_inherited_flags(argv) == [
+            "--profile=work",
+            "--model=anthropic/claude-sonnet-4",
+        ]
+
+    def test_skips_unknown_flags(self):
+        argv = ["--foo", "bar", "--tui"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["--tui"]
+
+    def test_does_not_consume_flag_like_value(self):
+        argv = ["--tui", "--resume", "abc123"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["--tui"]
+
+    def test_preserves_multiple_skills(self):
+        argv = ["-s", "foo", "-s", "bar", "--tui"]
+        assert relaunch_mod._extract_inherited_flags(argv) == ["-s", "foo", "-s", "bar", "--tui"]
+
+
+class TestInheritedFlagTable:
+    """Sanity-check the argparse-introspected table that drives extraction."""
+
+    def test_short_and_long_aliases_are_paired(self):
+        table = dict(relaunch_mod._INHERITED_FLAGS_TABLE)
+        # Each pair declared together in the parser shares takes_value.
+        for short, long_ in [
+            ("-p", "--profile"),
+            ("-m", "--model"),
+            ("-s", "--skills"),
+        ]:
+            assert table[short] == table[long_], f"{short}/{long_} disagree"
+
+    def test_store_true_flags_do_not_take_value(self):
+        table = dict(relaunch_mod._INHERITED_FLAGS_TABLE)
+        for flag in ["--tui", "--dev", "--yolo", "--ignore-user-config", "--ignore-rules"]:
+            assert table[flag] is False, f"{flag} should not take a value"
+
+    def test_value_flags_take_value(self):
+        table = dict(relaunch_mod._INHERITED_FLAGS_TABLE)
+        for flag in ["--profile", "--model", "--provider", "--skills"]:
+            assert table[flag] is True, f"{flag} should take a value"
+
+    def test_excluded_flags_are_not_inherited(self):
+        table = dict(relaunch_mod._INHERITED_FLAGS_TABLE)
+        # --worktree creates a new worktree per process; inheriting would
+        # orphan the parent's. Chat-only flags (--quiet/-Q, --verbose/-v,
+        # --source) can't be in argv at the existing relaunch callsites.
+        for flag in ["-w", "--worktree", "-Q", "--quiet", "-v", "--verbose", "--source"]:
+            assert flag not in table, f"{flag} should not be inherited"
+
+
+class TestBuildRelaunchArgv:
+    def test_uses_bin_when_available(self, monkeypatch):
+        monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: "/usr/bin/hermes")
+        argv = relaunch_mod.build_relaunch_argv(["--resume", "abc"])
+        assert argv[0] == "/usr/bin/hermes"
+
+    def test_falls_back_to_python_module(self, monkeypatch):
+        monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: None)
+        argv = relaunch_mod.build_relaunch_argv(["--resume", "abc"])
+        assert argv == [sys.executable, "-m", "hermes_cli.main", "--resume", "abc"]
+
+    def test_preserves_inherited_flags(self, monkeypatch):
+        monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: "/usr/bin/hermes")
+        original = ["--tui", "--dev", "--profile", "work", "sessions", "browse"]
+        argv = relaunch_mod.build_relaunch_argv(["--resume", "abc"], original_argv=original)
+        assert "--tui" in argv
+        assert "--dev" in argv
+        assert "--profile" in argv
+        assert "work" in argv
+        assert "--resume" in argv
+        assert "abc" in argv
+        # The original subcommand should not survive
+        assert "sessions" not in argv
+        assert "browse" not in argv
+
+    def test_can_disable_preserve(self, monkeypatch):
+        monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: "/usr/bin/hermes")
+        original = ["--tui", "chat"]
+        argv = relaunch_mod.build_relaunch_argv(
+            ["--resume", "abc"], preserve_inherited=False, original_argv=original
+        )
+        assert "--tui" not in argv
+        assert argv == ["/usr/bin/hermes", "--resume", "abc"]
+
+
+class TestRelaunch:
+    def test_calls_execvp(self, monkeypatch):
+        calls = []
+
+        def fake_execvp(path, argv):
+            calls.append((path, argv))
+            raise SystemExit(0)
+
+        monkeypatch.setattr(relaunch_mod.os, "execvp", fake_execvp)
+        monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: "/usr/bin/hermes")
+
+        with pytest.raises(SystemExit):
+            relaunch_mod.relaunch(["--resume", "abc"])
+
+        assert calls == [("/usr/bin/hermes", ["/usr/bin/hermes", "--resume", "abc"])]

tests/hermes_cli/test_runtime_provider_resolution.py

@@ -1998,6 +1998,7 @@ class TestAzureAnthropicEnvVarHint:
 
         assert resolved["api_key"] == "fallback-works"
 
+
     def test_no_key_anywhere_raises_helpful_error(self, monkeypatch):
         """When nothing resolves, the error message mentions key_env as an option."""
         monkeypatch.delenv("AZURE_ANTHROPIC_KEY", raising=False)
@@ -2168,3 +2169,67 @@ class TestTencentTokenhubRuntimeResolution:
         assert resolved["base_url"] == "https://explicit-proxy.example.com/v1"
         assert resolved["source"] == "explicit"
 
+# ---------------------------------------------------------------------------
+# minimax-oauth runtime resolution tests (added by feat/minimax-oauth-provider)
+# ---------------------------------------------------------------------------
+
+def test_minimax_oauth_runtime_returns_anthropic_messages_mode(monkeypatch):
+    """resolve_runtime_provider for minimax-oauth must return api_mode='anthropic_messages'."""
+    from hermes_cli.auth import MINIMAX_OAUTH_GLOBAL_INFERENCE
+
+    monkeypatch.setattr(rp, "resolve_provider", lambda *a, **k: "minimax-oauth")
+    monkeypatch.setattr(rp, "_get_model_config", lambda: {"provider": "minimax-oauth"})
+    monkeypatch.setattr(rp, "load_pool", lambda provider: None)
+    monkeypatch.setattr(
+        rp,
+        "_resolve_named_custom_runtime",
+        lambda **k: None,
+    )
+    monkeypatch.setattr(
+        rp,
+        "_resolve_explicit_runtime",
+        lambda **k: None,
+    )
+
+    fake_creds = {
+        "provider": "minimax-oauth",
+        "api_key": "mock-access-token",
+        "base_url": MINIMAX_OAUTH_GLOBAL_INFERENCE.rstrip("/"),
+        "source": "oauth",
+    }
+
+    import hermes_cli.auth as auth_mod
+    monkeypatch.setattr(auth_mod, "resolve_minimax_oauth_runtime_credentials",
+                        lambda **k: fake_creds)
+
+    resolved = rp.resolve_runtime_provider(requested="minimax-oauth")
+
+    assert resolved["provider"] == "minimax-oauth"
+    assert resolved["api_mode"] == "anthropic_messages"
+    assert resolved["api_key"] == "mock-access-token"
+
+
+def test_minimax_oauth_runtime_uses_inference_base_url(monkeypatch):
+    """Base URL returned by resolve_runtime_provider should match the OAuth credentials."""
+    from hermes_cli.auth import MINIMAX_OAUTH_CN_INFERENCE
+
+    monkeypatch.setattr(rp, "resolve_provider", lambda *a, **k: "minimax-oauth")
+    monkeypatch.setattr(rp, "_get_model_config", lambda: {"provider": "minimax-oauth"})
+    monkeypatch.setattr(rp, "load_pool", lambda provider: None)
+    monkeypatch.setattr(rp, "_resolve_named_custom_runtime", lambda **k: None)
+    monkeypatch.setattr(rp, "_resolve_explicit_runtime", lambda **k: None)
+
+    fake_creds = {
+        "provider": "minimax-oauth",
+        "api_key": "cn-token",
+        "base_url": MINIMAX_OAUTH_CN_INFERENCE.rstrip("/"),
+        "source": "oauth",
+    }
+
+    import hermes_cli.auth as auth_mod
+    monkeypatch.setattr(auth_mod, "resolve_minimax_oauth_runtime_credentials",
+                        lambda **k: fake_creds)
+
+    resolved = rp.resolve_runtime_provider(requested="minimax-oauth")
+
+    assert MINIMAX_OAUTH_CN_INFERENCE.rstrip("/") in resolved["base_url"]

tests/hermes_cli/test_set_config_value.py

@@ -127,6 +127,13 @@ class TestConfigYamlRouting:
             or "TERMINAL_DOCKER_MOUNT_CWD_TO_WORKSPACE=True" in env_content
         )
 
+    def test_terminal_vercel_runtime_goes_to_config_and_env(self, _isolated_hermes_home):
+        set_config_value("terminal.vercel_runtime", "python3.13")
+        config = _read_config(_isolated_hermes_home)
+        env_content = _read_env(_isolated_hermes_home)
+        assert "vercel_runtime: python3.13" in config
+        assert "TERMINAL_VERCEL_RUNTIME=python3.13" in env_content
+
 
 # ---------------------------------------------------------------------------
 # Empty / falsy values — regression tests for #4277

tests/hermes_cli/test_setup.py

@@ -1,5 +1,6 @@
 """Tests for setup.py configuration flows."""
 import json
+import os
 import sys
 import types
 
@@ -29,6 +30,17 @@ def _clear_provider_env(monkeypatch):
         monkeypatch.delenv(key, raising=False)
 
 
+def _clear_vercel_env(monkeypatch):
+    for key in (
+        "TERMINAL_VERCEL_RUNTIME",
+        "VERCEL_OIDC_TOKEN",
+        "VERCEL_TOKEN",
+        "VERCEL_PROJECT_ID",
+        "VERCEL_TEAM_ID",
+    ):
+        monkeypatch.delenv(key, raising=False)
+
+
 def _stub_tts(monkeypatch):
     """Stub out TTS prompts so setup_model_provider doesn't block."""
     monkeypatch.setattr("hermes_cli.setup.prompt_choice", lambda q, c, d=0: (
@@ -162,12 +174,13 @@ def test_setup_gateway_skips_service_install_when_systemctl_missing(monkeypatch,
         "WEBHOOK_ENABLED": "",
     }
 
+    import hermes_cli.gateway as gateway_mod
+
     monkeypatch.setattr(setup_mod, "get_env_value", lambda key: env.get(key, ""))
+    monkeypatch.setattr(gateway_mod, "get_env_value", lambda key: env.get(key, ""))
     monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *args, **kwargs: False)
     monkeypatch.setattr("platform.system", lambda: "Linux")
 
-    import hermes_cli.gateway as gateway_mod
-
     monkeypatch.setattr(gateway_mod, "supports_systemd_services", lambda: False)
     monkeypatch.setattr(gateway_mod, "is_macos", lambda: False)
     monkeypatch.setattr(gateway_mod, "_is_service_installed", lambda: False)
@@ -200,12 +213,13 @@ def test_setup_gateway_in_container_shows_docker_guidance(monkeypatch, capsys):
         "WEBHOOK_ENABLED": "",
     }
 
+    import hermes_cli.gateway as gateway_mod
+
     monkeypatch.setattr(setup_mod, "get_env_value", lambda key: env.get(key, ""))
+    monkeypatch.setattr(gateway_mod, "get_env_value", lambda key: env.get(key, ""))
     monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *args, **kwargs: False)
     monkeypatch.setattr("platform.system", lambda: "Linux")
 
-    import hermes_cli.gateway as gateway_mod
-
     monkeypatch.setattr(gateway_mod, "supports_systemd_services", lambda: False)
     monkeypatch.setattr(gateway_mod, "is_macos", lambda: False)
     monkeypatch.setattr(gateway_mod, "_is_service_installed", lambda: False)
@@ -480,28 +494,91 @@ def test_modal_setup_persists_direct_mode_when_user_chooses_their_own_account(tm
     assert config["terminal"]["modal_mode"] == "direct"
 
 
-def test_resolve_hermes_chat_argv_prefers_which(monkeypatch):
-    from hermes_cli import setup as setup_mod
+def test_vercel_setup_configures_access_token_auth(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    _clear_vercel_env(monkeypatch)
+    monkeypatch.setenv("VERCEL_OIDC_TOKEN", "old-oidc")
+    monkeypatch.setitem(sys.modules, "vercel", types.ModuleType("vercel"))
+    config = load_config()
 
-    monkeypatch.setattr(setup_mod.shutil, "which", lambda name: "/usr/local/bin/hermes" if name == "hermes" else None)
+    def fake_prompt_choice(question, choices, default=0):
+        if question == "Select terminal backend:":
+            return 5
+        raise AssertionError(f"Unexpected prompt_choice call: {question}")
 
-    assert setup_mod._resolve_hermes_chat_argv() == ["/usr/local/bin/hermes", "chat"]
+    prompt_values = iter(["python3.13", "yes", "2", "4096", "token", "project", "team"])
 
+    monkeypatch.setattr("hermes_cli.setup.prompt_choice", fake_prompt_choice)
+    monkeypatch.setattr("hermes_cli.setup.prompt", lambda *args, **kwargs: next(prompt_values))
 
-def test_resolve_hermes_chat_argv_falls_back_to_module(monkeypatch):
-    from hermes_cli import setup as setup_mod
+    from hermes_cli.setup import setup_terminal_backend
 
-    monkeypatch.setattr(setup_mod.shutil, "which", lambda _name: None)
-    monkeypatch.setattr(setup_mod.importlib.util, "find_spec", lambda name: object() if name == "hermes_cli" else None)
+    setup_terminal_backend(config)
 
-    assert setup_mod._resolve_hermes_chat_argv() == [sys.executable, "-m", "hermes_cli.main", "chat"]
+    assert config["terminal"]["backend"] == "vercel_sandbox"
+    assert config["terminal"]["vercel_runtime"] == "python3.13"
+    assert config["terminal"]["container_disk"] == 51200
+    assert os.environ["TERMINAL_VERCEL_RUNTIME"] == "python3.13"
+    assert "VERCEL_OIDC_TOKEN" not in os.environ
+    assert os.environ["VERCEL_TOKEN"] == "token"
+    assert os.environ["VERCEL_PROJECT_ID"] == "project"
+    assert os.environ["VERCEL_TEAM_ID"] == "team"
 
 
-def test_offer_launch_chat_execs_fresh_process(monkeypatch):
+def test_vercel_setup_prefills_project_and_team_from_link_file(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    _clear_vercel_env(monkeypatch)
+    project_root = tmp_path / "project"
+    nested = project_root / "app" / "src"
+    nested.mkdir(parents=True)
+    vercel_dir = project_root / ".vercel"
+    vercel_dir.mkdir()
+    (vercel_dir / "project.json").write_text(
+        json.dumps({"projectId": "linked-project", "orgId": "linked-team"}),
+        encoding="utf-8",
+    )
+    monkeypatch.chdir(nested)
+    monkeypatch.setitem(sys.modules, "vercel", types.ModuleType("vercel"))
+    config = load_config()
+    config["terminal"]["container_disk"] = 999
+
+    def fake_prompt_choice(question, choices, default=0):
+        if question == "Select terminal backend:":
+            return 5
+        raise AssertionError(f"Unexpected prompt_choice call: {question}")
+
+    prompt_values = iter(["node24", "no", "1", "5120", "token", "", ""])
+    defaults = {}
+
+    def fake_prompt(message, default="", **kwargs):
+        defaults[message] = default
+        value = next(prompt_values)
+        return value or default
+
+    monkeypatch.setattr("hermes_cli.setup.prompt_choice", fake_prompt_choice)
+    monkeypatch.setattr("hermes_cli.setup.prompt", fake_prompt)
+
+    from hermes_cli.setup import setup_terminal_backend
+
+    setup_terminal_backend(config)
+
+    assert config["terminal"]["backend"] == "vercel_sandbox"
+    assert config["terminal"]["container_persistent"] is False
+    assert config["terminal"]["container_disk"] == 51200
+    assert "VERCEL_OIDC_TOKEN" not in os.environ
+    assert os.environ["VERCEL_TOKEN"] == "token"
+    assert os.environ["VERCEL_PROJECT_ID"] == "linked-project"
+    assert os.environ["VERCEL_TEAM_ID"] == "linked-team"
+    assert defaults["    Vercel project ID"] == "linked-project"
+    assert defaults["    Vercel team ID"] == "linked-team"
+
+
+def test_offer_launch_chat_relaunches_via_bin(monkeypatch):
     from hermes_cli import setup as setup_mod
+    from hermes_cli import relaunch as relaunch_mod
 
     monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *_args, **_kwargs: True)
-    monkeypatch.setattr(setup_mod, "_resolve_hermes_chat_argv", lambda: ["/usr/local/bin/hermes", "chat"])
+    monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: "/usr/local/bin/hermes")
 
     exec_calls = []
 
@@ -509,7 +586,7 @@ def test_offer_launch_chat_execs_fresh_process(monkeypatch):
         exec_calls.append((path, argv))
         raise SystemExit(0)
 
-    monkeypatch.setattr(setup_mod.os, "execvp", fake_execvp)
+    monkeypatch.setattr(relaunch_mod.os, "execvp", fake_execvp)
 
     with pytest.raises(SystemExit):
         setup_mod._offer_launch_chat()
@@ -517,13 +594,22 @@ def test_offer_launch_chat_execs_fresh_process(monkeypatch):
     assert exec_calls == [("/usr/local/bin/hermes", ["/usr/local/bin/hermes", "chat"])]
 
 
-def test_offer_launch_chat_manual_fallback_when_unresolvable(monkeypatch, capsys):
+def test_offer_launch_chat_falls_back_to_module(monkeypatch):
     from hermes_cli import setup as setup_mod
+    from hermes_cli import relaunch as relaunch_mod
 
     monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *_args, **_kwargs: True)
-    monkeypatch.setattr(setup_mod, "_resolve_hermes_chat_argv", lambda: None)
+    monkeypatch.setattr(relaunch_mod, "resolve_hermes_bin", lambda: None)
 
-    setup_mod._offer_launch_chat()
+    exec_calls = []
 
-    captured = capsys.readouterr()
-    assert "Run 'hermes chat' manually" in captured.out
+    def fake_execvp(path, argv):
+        exec_calls.append((path, argv))
+        raise SystemExit(0)
+
+    monkeypatch.setattr(relaunch_mod.os, "execvp", fake_execvp)
+
+    with pytest.raises(SystemExit):
+        setup_mod._offer_launch_chat()
+
+    assert exec_calls == [(sys.executable, [sys.executable, "-m", "hermes_cli.main", "chat"])]

tests/hermes_cli/test_setup_irc.py

@@ -0,0 +1,245 @@
+"""Tests for IRC gateway configuration via `hermes setup gateway` UI.
+
+Covers the full plugin-platform discovery → status → configure flow so that
+a fresh Hermes install (no state, no env vars) can set up IRC through the
+interactive setup menus.
+"""
+
+import os
+import pytest
+
+from gateway.platform_registry import PlatformEntry, platform_registry
+
+
+def _register_irc_platform(**overrides):
+    """Manually register the IRC platform entry as if discover_plugins() found it.
+
+    Tests run outside the normal plugin-discovery path, so we inject the entry
+    directly into the singleton registry and yield its dict shape.
+    """
+    defaults = dict(
+        name="irc",
+        label="IRC",
+        adapter_factory=lambda cfg: None,
+        check_fn=lambda: bool(os.getenv("IRC_SERVER", "") and os.getenv("IRC_CHANNEL", "")),
+        validate_config=None,
+        required_env=["IRC_SERVER", "IRC_CHANNEL", "IRC_NICKNAME"],
+        install_hint="No extra packages needed (stdlib only)",
+        setup_fn=lambda: None,
+        source="plugin",
+        plugin_name="irc_platform",
+        allowed_users_env="IRC_ALLOWED_USERS",
+        allow_all_env="IRC_ALLOW_ALL_USERS",
+        max_message_length=450,
+        pii_safe=False,
+        emoji="💬",
+        allow_update_command=True,
+        platform_hint="You are chatting via IRC.",
+    )
+    defaults.update(overrides)
+    entry = PlatformEntry(**defaults)
+    platform_registry.register(entry)
+    return {
+        "key": entry.name,
+        "label": entry.label,
+        "emoji": entry.emoji,
+        "token_var": entry.required_env[0] if entry.required_env else "",
+        "install_hint": entry.install_hint,
+        "_registry_entry": entry,
+    }
+
+
+def _unregister_irc_platform():
+    platform_registry.unregister("irc")
+
+
+# ── Fresh-install discovery ─────────────────────────────────────────────────
+
+
+class TestIRCFreshInstallDiscovery:
+    """IRC appears in the setup menu on a brand-new Hermes install."""
+
+    def test_irc_appears_in_all_platforms(self, monkeypatch):
+        """When the IRC plugin is registered, _all_platforms() surfaces it."""
+        import hermes_cli.gateway as gateway_mod
+
+        _register_irc_platform()
+        try:
+            # Ensure no stale env vars leak in
+            for key in ("IRC_SERVER", "IRC_CHANNEL", "IRC_NICKNAME"):
+                monkeypatch.delenv(key, raising=False)
+
+            platforms = gateway_mod._all_platforms()
+            keys = {p["key"] for p in platforms}
+            assert "irc" in keys
+
+            irc_plat = next(p for p in platforms if p["key"] == "irc")
+            assert irc_plat["label"] == "IRC"
+            assert irc_plat["emoji"] == "💬"
+        finally:
+            _unregister_irc_platform()
+
+    def test_irc_status_not_configured_when_fresh(self, monkeypatch):
+        """On a fresh install with no env vars, IRC shows 'not configured'."""
+        import hermes_cli.gateway as gateway_mod
+
+        plat = _register_irc_platform()
+        try:
+            for key in ("IRC_SERVER", "IRC_CHANNEL", "IRC_NICKNAME"):
+                monkeypatch.delenv(key, raising=False)
+
+            status = gateway_mod._platform_status(plat)
+            assert status == "not configured"
+        finally:
+            _unregister_irc_platform()
+
+    def test_irc_status_configured_when_env_set(self, monkeypatch):
+        """After the user sets IRC_SERVER and IRC_CHANNEL, status is 'configured'."""
+        import hermes_cli.gateway as gateway_mod
+
+        plat = _register_irc_platform()
+        try:
+            monkeypatch.setenv("IRC_SERVER", "irc.libera.chat")
+            monkeypatch.setenv("IRC_CHANNEL", "#hermes")
+            monkeypatch.setenv("IRC_NICKNAME", "hermes-bot")
+
+            status = gateway_mod._platform_status(plat)
+            assert status == "configured"
+        finally:
+            _unregister_irc_platform()
+
+    def test_irc_status_partial_when_only_server_set(self, monkeypatch):
+        """If only IRC_SERVER is set, the platform is still not configured."""
+        import hermes_cli.gateway as gateway_mod
+
+        plat = _register_irc_platform()
+        try:
+            monkeypatch.delenv("IRC_CHANNEL", raising=False)
+            monkeypatch.delenv("IRC_NICKNAME", raising=False)
+            monkeypatch.setenv("IRC_SERVER", "irc.libera.chat")
+
+            status = gateway_mod._platform_status(plat)
+            assert status == "not configured"
+        finally:
+            _unregister_irc_platform()
+
+
+# ── Interactive setup dispatch ──────────────────────────────────────────────
+
+
+class TestIRCInteractiveSetup:
+    """The setup UI dispatches to IRC's interactive_setup() correctly."""
+
+    def test_configure_platform_dispatches_to_irc_setup_fn(self, monkeypatch, capsys):
+        """_configure_platform() calls the IRC plugin's setup_fn when selected."""
+        import hermes_cli.gateway as gateway_mod
+
+        calls = []
+
+        def fake_setup():
+            calls.append("setup_called")
+            print("IRC setup complete!")
+
+        plat = _register_irc_platform(setup_fn=fake_setup)
+        try:
+            gateway_mod._configure_platform(plat)
+        finally:
+            _unregister_irc_platform()
+
+        assert "setup_called" in calls
+        out = capsys.readouterr().out
+        assert "IRC setup complete!" in out
+
+
+    def test_configure_platform_fallback_when_no_setup_fn(self, monkeypatch, capsys):
+        """A plugin with no setup_fn falls back to env-var instructions."""
+        import hermes_cli.gateway as gateway_mod
+
+        plat = _register_irc_platform(setup_fn=None)
+        try:
+            gateway_mod._configure_platform(plat)
+        finally:
+            _unregister_irc_platform()
+
+        out = capsys.readouterr().out
+        assert "IRC" in out
+        assert "IRC_SERVER" in out
+
+
+# ── End-to-end fresh-install gateway setup ──────────────────────────────────
+
+
+class TestIRCGatewaySetupFreshInstall:
+    """Simulate the full `hermes setup gateway` experience with IRC present."""
+
+    def test_setup_gateway_shows_irc_in_platform_menu(self, monkeypatch, capsys, tmp_path):
+        """The gateway setup menu lists IRC among the available platforms."""
+        import hermes_cli.gateway as gateway_mod
+        from hermes_cli import setup as setup_mod
+
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+        _register_irc_platform()
+        try:
+            for key in ("IRC_SERVER", "IRC_CHANNEL", "IRC_NICKNAME"):
+                monkeypatch.delenv(key, raising=False)
+
+            # Sanity-check: IRC must be visible to _all_platforms()
+            platforms = gateway_mod._all_platforms()
+            assert any(p["key"] == "irc" for p in platforms), \
+                f"IRC not in platforms: {[p['key'] for p in platforms]}"
+
+            # Capture what prompt_checklist is asked to display
+            checklist_calls = []
+
+            def capture_prompt_checklist(question, choices, pre_selected=None):
+                checklist_calls.append({"question": question, "choices": choices})
+                return []  # nothing selected → clean exit
+
+            monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *a, **kw: False)
+            monkeypatch.setattr(setup_mod, "prompt_checklist", capture_prompt_checklist)
+            monkeypatch.setattr(gateway_mod, "supports_systemd_services", lambda: False)
+            monkeypatch.setattr(gateway_mod, "is_macos", lambda: False)
+            monkeypatch.setattr(gateway_mod, "_is_service_installed", lambda: False)
+            monkeypatch.setattr(gateway_mod, "_is_service_running", lambda: False)
+
+            setup_mod.setup_gateway({})
+
+            # Find the platform-selection prompt
+            platform_prompt = next(
+                (c for c in checklist_calls if "platform" in c["question"].lower()),
+                None,
+            )
+            assert platform_prompt is not None, \
+                f"No platform prompt found in {checklist_calls}"
+            choices_text = "\n".join(platform_prompt["choices"])
+            assert "IRC" in choices_text
+            assert "💬" in choices_text
+            assert "not configured" in choices_text.lower()
+        finally:
+            _unregister_irc_platform()
+
+    def test_setup_gateway_irc_counts_as_messaging_platform(self, monkeypatch, capsys, tmp_path):
+        """When IRC is configured, setup_gateway counts it as a messaging platform."""
+        import hermes_cli.gateway as gateway_mod
+        from hermes_cli import setup as setup_mod
+
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+        _register_irc_platform()
+        try:
+            monkeypatch.setenv("IRC_SERVER", "irc.libera.chat")
+            monkeypatch.setenv("IRC_CHANNEL", "#hermes")
+            monkeypatch.setenv("IRC_NICKNAME", "hermes-bot")
+
+            monkeypatch.setattr(setup_mod, "prompt_yes_no", lambda *a, **kw: False)
+            monkeypatch.setattr(setup_mod, "prompt_choice", lambda *a, **kw: 0)
+            monkeypatch.setattr(gateway_mod, "supports_systemd_services", lambda: False)
+            monkeypatch.setattr(gateway_mod, "is_macos", lambda: False)
+            monkeypatch.setattr(gateway_mod, "_is_service_installed", lambda: False)
+            monkeypatch.setattr(gateway_mod, "_is_service_running", lambda: False)
+
+            setup_mod.setup_gateway({})
+
+            out = capsys.readouterr().out
+            assert "Messaging platforms configured!" in out
+        finally:
+            _unregister_irc_platform()

tests/hermes_cli/test_setup_openclaw_migration.py

@@ -419,7 +419,12 @@ class TestGetSectionConfigSummary:
                 return "disc456"
             return ""
 
-        with patch.object(setup_mod, "get_env_value", side_effect=env_side):
+        # Also patch gateway module's binding since _platform_status()
+        # reads from hermes_cli.gateway.get_env_value after the setup
+        # flows were unified via platform_registry.
+        import hermes_cli.gateway as gateway_mod
+        with patch.object(setup_mod, "get_env_value", side_effect=env_side), \
+             patch.object(gateway_mod, "get_env_value", side_effect=env_side):
             result = setup_mod._get_section_config_summary({}, "gateway")
         assert "Telegram" in result
         assert "Discord" in result
@@ -471,7 +476,9 @@ class TestGetSectionConfigSummary:
         def env_side(key):
             return "true" if key == "WHATSAPP_ENABLED" else ""
 
-        with patch.object(setup_mod, "get_env_value", side_effect=env_side):
+        import hermes_cli.gateway as gateway_mod
+        with patch.object(setup_mod, "get_env_value", side_effect=env_side), \
+             patch.object(gateway_mod, "get_env_value", side_effect=env_side):
             result = setup_mod._get_section_config_summary({}, "gateway")
         assert result is not None
         assert "WhatsApp" in result
@@ -481,7 +488,9 @@ class TestGetSectionConfigSummary:
         def env_side(key):
             return "http://signal.local" if key == "SIGNAL_HTTP_URL" else ""
 
-        with patch.object(setup_mod, "get_env_value", side_effect=env_side):
+        import hermes_cli.gateway as gateway_mod
+        with patch.object(setup_mod, "get_env_value", side_effect=env_side), \
+             patch.object(gateway_mod, "get_env_value", side_effect=env_side):
             result = setup_mod._get_section_config_summary({}, "gateway")
         assert result is not None
         assert "Signal" in result
@@ -529,13 +538,28 @@ class TestGetSectionConfigSummary:
         assert result == "gpt-5"
 
     def test_gateway_matches_platform_registry(self):
-        """Every platform in _GATEWAY_PLATFORMS should be recognised by its
-        own env-var sentinel — i.e. the summary must not drift from the
+        """Every built-in platform should be recognised by its primary
+        env-var sentinel — i.e. the summary must not drift from the
         registry used by the setup checklist."""
-        for label, env_var, _fn in setup_mod._GATEWAY_PLATFORMS:
+        from hermes_cli.gateway import _PLATFORMS
+
+        for plat in _PLATFORMS:
+            label = plat["label"]
+            env_var = plat.get("token_var")
+            if not env_var:
+                continue
+            # Some platforms require a specific value shape (e.g. WhatsApp
+            # needs the literal "true"). Use a sentinel that satisfies every
+            # real validator _platform_status() currently checks.
             def env_side(key, _target=env_var):
-                return "x" if key == _target else ""
-            with patch.object(setup_mod, "get_env_value", side_effect=env_side):
+                if key != _target:
+                    return ""
+                if _target == "WHATSAPP_ENABLED":
+                    return "true"
+                return "x"
+            import hermes_cli.gateway as gateway_mod
+            with patch.object(setup_mod, "get_env_value", side_effect=env_side), \
+                 patch.object(gateway_mod, "get_env_value", side_effect=env_side):
                 result = setup_mod._get_section_config_summary({}, "gateway")
             expected = setup_mod._gateway_platform_short_label(label)
             assert result is not None, f"{label} ({env_var}) not recognised"

tests/hermes_cli/test_status.py

@@ -79,3 +79,33 @@ def test_show_status_reports_nous_auth_error(monkeypatch, capsys, tmp_path):
     assert "Error:      Refresh session has been revoked" in output
     assert "Access exp:" in output
     assert "Key exp:" in output
+
+
+def test_show_status_reports_vercel_backend_contract(monkeypatch, capsys, tmp_path):
+    from hermes_cli import status as status_mod
+    import hermes_cli.auth as auth_mod
+    import hermes_cli.gateway as gateway_mod
+
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    monkeypatch.setenv("TERMINAL_ENV", "vercel_sandbox")
+    monkeypatch.setenv("TERMINAL_VERCEL_RUNTIME", "python3.13")
+    monkeypatch.setenv("TERMINAL_CONTAINER_PERSISTENT", "true")
+    monkeypatch.setenv("VERCEL_OIDC_TOKEN", "oidc-token")
+    monkeypatch.setattr(status_mod.importlib.util, "find_spec", lambda name: object() if name == "vercel" else None)
+    monkeypatch.setattr(status_mod, "load_config", lambda: {"terminal": {"backend": "vercel_sandbox"}}, raising=False)
+    monkeypatch.setattr(auth_mod, "get_nous_auth_status", lambda: {}, raising=False)
+    monkeypatch.setattr(auth_mod, "get_codex_auth_status", lambda: {}, raising=False)
+    monkeypatch.setattr(auth_mod, "get_qwen_auth_status", lambda: {}, raising=False)
+    monkeypatch.setattr(gateway_mod, "find_gateway_pids", lambda exclude_pids=None: [], raising=False)
+
+    status_mod.show_status(SimpleNamespace(all=False, deep=False))
+
+    output = capsys.readouterr().out
+    assert "Backend:      vercel_sandbox" in output
+    assert "Runtime:      python3.13" in output
+    assert "Auth:" in output and "OIDC token via VERCEL_OIDC_TOKEN" in output
+    assert "Auth detail:  mode: OIDC" in output
+    assert "Auth detail:  active env: VERCEL_OIDC_TOKEN" in output
+    assert "oidc-token" not in output
+    assert "snapshot filesystem" in output
+    assert "live processes do not survive" in output

tests/hermes_cli/test_tui_resume_flow.py

@@ -12,6 +12,7 @@ def _args(**overrides):
         "model": None,
         "provider": None,
         "resume": None,
+        "toolsets": None,
         "tui": True,
         "tui_dev": False,
     }
@@ -35,7 +36,7 @@ def test_cmd_chat_tui_continue_uses_latest_tui_session(monkeypatch, main_mod):
         calls.append(source)
         return "20260408_235959_a1b2c3" if source == "tui" else None
 
-    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None):
+    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None, toolsets=None):
         captured["resume"] = resume_session_id
         raise SystemExit(0)
 
@@ -62,7 +63,7 @@ def test_cmd_chat_tui_continue_falls_back_to_latest_cli_session(monkeypatch, mai
             return "20260408_235959_d4e5f6"
         return None
 
-    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None):
+    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None, toolsets=None):
         captured["resume"] = resume_session_id
         raise SystemExit(0)
 
@@ -80,7 +81,7 @@ def test_cmd_chat_tui_continue_falls_back_to_latest_cli_session(monkeypatch, mai
 def test_cmd_chat_tui_resume_resolves_title_before_launch(monkeypatch, main_mod):
     captured = {}
 
-    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None):
+    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None, toolsets=None):
         captured["resume"] = resume_session_id
         raise SystemExit(0)
 
@@ -98,12 +99,13 @@ def test_cmd_chat_tui_resume_resolves_title_before_launch(monkeypatch, main_mod)
 def test_cmd_chat_tui_passes_model_and_provider(monkeypatch, main_mod):
     captured = {}
 
-    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None):
+    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None, toolsets=None):
         captured.update(
             {
                 "model": model,
                 "provider": provider,
                 "resume": resume_session_id,
+                "toolsets": toolsets,
                 "tui_dev": tui_dev,
             }
         )
@@ -120,11 +122,193 @@ def test_cmd_chat_tui_passes_model_and_provider(monkeypatch, main_mod):
         "model": "anthropic/claude-sonnet-4.6",
         "provider": "anthropic",
         "resume": None,
+        "toolsets": None,
         "tui_dev": False,
     }
 
 
-def test_launch_tui_exports_model_and_provider(monkeypatch, main_mod):
+def test_cmd_chat_tui_passes_toolsets(monkeypatch, main_mod):
+    captured = {}
+
+    def fake_launch(resume_session_id=None, tui_dev=False, model=None, provider=None, toolsets=None):
+        captured["toolsets"] = toolsets
+        raise SystemExit(0)
+
+    monkeypatch.setattr(main_mod, "_launch_tui", fake_launch)
+
+    with pytest.raises(SystemExit):
+        main_mod.cmd_chat(_args(toolsets="web,terminal"))
+
+    assert captured["toolsets"] == "web,terminal"
+
+
+def test_main_top_level_tui_accepts_toolsets(monkeypatch, main_mod):
+    captured = {}
+
+    import hermes_cli.config as config_mod
+
+    monkeypatch.setattr(sys, "argv", ["hermes", "--tui", "--toolsets", "web,terminal"])
+    monkeypatch.setitem(sys.modules, "hermes_cli.plugins", types.SimpleNamespace(discover_plugins=lambda: None))
+    monkeypatch.setitem(sys.modules, "tools.mcp_tool", types.SimpleNamespace(discover_mcp_tools=lambda: None))
+    monkeypatch.setattr(config_mod, "load_config", lambda: {})
+    monkeypatch.setattr(config_mod, "get_container_exec_info", lambda: None)
+    monkeypatch.setitem(
+        sys.modules,
+        "agent.shell_hooks",
+        types.SimpleNamespace(register_from_config=lambda _cfg, accept_hooks=False: None),
+    )
+    monkeypatch.setattr(main_mod, "cmd_chat", lambda args: captured.update({"toolsets": args.toolsets, "tui": args.tui}))
+
+    main_mod.main()
+
+    assert captured == {"toolsets": "web,terminal", "tui": True}
+
+
+def test_main_top_level_oneshot_accepts_toolsets(monkeypatch, main_mod):
+    captured = {}
+
+    import hermes_cli.config as config_mod
+
+    monkeypatch.setattr(sys, "argv", ["hermes", "-z", "hello", "--toolsets", "web,terminal"])
+    monkeypatch.setitem(sys.modules, "hermes_cli.plugins", types.SimpleNamespace(discover_plugins=lambda: None))
+    monkeypatch.setitem(sys.modules, "tools.mcp_tool", types.SimpleNamespace(discover_mcp_tools=lambda: None))
+    monkeypatch.setattr(config_mod, "load_config", lambda: {})
+    monkeypatch.setattr(config_mod, "get_container_exec_info", lambda: None)
+    monkeypatch.setitem(
+        sys.modules,
+        "agent.shell_hooks",
+        types.SimpleNamespace(register_from_config=lambda _cfg, accept_hooks=False: None),
+    )
+    monkeypatch.setitem(
+        sys.modules,
+        "hermes_cli.oneshot",
+        types.SimpleNamespace(run_oneshot=lambda prompt, **kwargs: captured.update({"prompt": prompt, **kwargs}) or 0),
+    )
+
+    with pytest.raises(SystemExit) as exc:
+        main_mod.main()
+
+    assert exc.value.code == 0
+    assert captured == {"prompt": "hello", "model": None, "provider": None, "toolsets": "web,terminal"}
+
+
+def _stub_plugin_discovery(monkeypatch):
+    monkeypatch.setitem(
+        sys.modules,
+        "hermes_cli.plugins",
+        types.SimpleNamespace(discover_plugins=lambda: None),
+    )
+
+
+def test_oneshot_rejects_invalid_only_toolsets(monkeypatch, capsys):
+    _stub_plugin_discovery(monkeypatch)
+    from hermes_cli.oneshot import run_oneshot
+
+    assert run_oneshot("hello", toolsets="nope") == 2
+    err = capsys.readouterr().err
+    assert "nope" in err
+    assert "did not contain any valid toolsets" in err
+
+
+def test_oneshot_filters_invalid_toolsets_before_redirect(monkeypatch, capsys):
+    _stub_plugin_discovery(monkeypatch)
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    valid, error = _validate_explicit_toolsets("web,nope")
+
+    assert valid == ["web"]
+    assert error is None
+    assert "nope" in capsys.readouterr().err
+
+
+def test_oneshot_all_toolsets_means_all_not_configured_cli():
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    valid, error = _validate_explicit_toolsets("all")
+
+    assert valid is None
+    assert error is None
+
+
+def test_oneshot_all_toolsets_warns_about_ignored_extra_entries(monkeypatch, capsys):
+    _stub_plugin_discovery(monkeypatch)
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    valid, error = _validate_explicit_toolsets("all,nope")
+
+    assert valid is None
+    assert error is None
+    assert "ignoring additional entries: nope" in capsys.readouterr().err
+
+
+def test_oneshot_accepts_plugin_toolset_after_discovery(monkeypatch):
+    import toolsets
+
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    discovered = {"ready": False}
+    original_validate = toolsets.validate_toolset
+
+    def fake_validate(name):
+        return name == "plugin_demo" and discovered["ready"] or original_validate(name)
+
+    monkeypatch.setattr(toolsets, "validate_toolset", fake_validate)
+    monkeypatch.setitem(
+        sys.modules,
+        "hermes_cli.plugins",
+        types.SimpleNamespace(discover_plugins=lambda: discovered.update({"ready": True})),
+    )
+
+    valid, error = _validate_explicit_toolsets("plugin_demo")
+
+    assert valid == ["plugin_demo"]
+    assert error is None
+
+
+def test_oneshot_rejects_disabled_mcp_toolset(monkeypatch, capsys):
+    _stub_plugin_discovery(monkeypatch)
+    import hermes_cli.config as config_mod
+
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    monkeypatch.setattr(
+        config_mod,
+        "read_raw_config",
+        lambda: {"mcp_servers": {"mcp-off": {"enabled": False}}},
+    )
+
+    valid, error = _validate_explicit_toolsets("mcp-off")
+
+    assert valid is None
+    assert error == "hermes -z: --toolsets did not contain any valid toolsets.\n"
+    err = capsys.readouterr().err
+    assert "ignoring disabled MCP servers" in err
+    assert "mcp-off" in err
+
+
+def test_oneshot_distinguishes_disabled_mcp_from_unknown(monkeypatch, capsys):
+    _stub_plugin_discovery(monkeypatch)
+    import hermes_cli.config as config_mod
+
+    from hermes_cli.oneshot import _validate_explicit_toolsets
+
+    monkeypatch.setattr(
+        config_mod,
+        "read_raw_config",
+        lambda: {"mcp_servers": {"mcp-off": {"enabled": False}}},
+    )
+
+    valid, error = _validate_explicit_toolsets("web,mcp-off,nope")
+
+    assert valid == ["web"]
+    assert error is None
+    err = capsys.readouterr().err
+    assert "ignoring unknown --toolsets entries: nope" in err
+    assert "ignoring disabled MCP servers" in err
+    assert "mcp-off" in err
+
+
+def test_launch_tui_exports_model_provider_and_toolsets(monkeypatch, main_mod):
     captured = {}
     active_path_during_call = None
 
@@ -144,13 +328,14 @@ def test_launch_tui_exports_model_and_provider(monkeypatch, main_mod):
     monkeypatch.setattr(main_mod.subprocess, "call", fake_call)
 
     with pytest.raises(SystemExit):
-        main_mod._launch_tui(model="nous/hermes-test", provider="nous")
+        main_mod._launch_tui(model="nous/hermes-test", provider="nous", toolsets="web, terminal")
 
     env = captured["env"]
     assert env["HERMES_MODEL"] == "nous/hermes-test"
     assert env["HERMES_INFERENCE_MODEL"] == "nous/hermes-test"
     assert env["HERMES_TUI_PROVIDER"] == "nous"
     assert env["HERMES_INFERENCE_PROVIDER"] == "nous"
+    assert env["HERMES_TUI_TOOLSETS"] == "web,terminal"
     active_path = Path(env["HERMES_TUI_ACTIVE_SESSION_FILE"])
     assert active_path.name.startswith("hermes-tui-active-session-")
     assert active_path.suffix == ".json"

tests/hermes_cli/test_update_autostash.py

@@ -333,7 +333,10 @@ def test_cmd_update_retries_optional_extras_individually_when_all_fails(monkeypa
             raise CalledProcessError(returncode=1, cmd=cmd)
         if cmd == ["/usr/bin/uv", "pip", "install", "-e", ".[mcp]", "--quiet"]:
             return SimpleNamespace(returncode=0)
-        return SimpleNamespace(returncode=0)
+        # Catch-all must include stdout/stderr so consumers that parse
+        # output (e.g. the dashboard-restart `ps -A` scan added in the
+        # updater) don't crash on AttributeError.
+        return SimpleNamespace(returncode=0, stdout="", stderr="")
 
     monkeypatch.setattr(hermes_main.subprocess, "run", fake_run)
 
@@ -370,7 +373,7 @@ def test_cmd_update_succeeds_with_extras(monkeypatch, tmp_path):
             return SimpleNamespace(stdout="1\n", stderr="", returncode=0)
         if cmd == ["git", "pull", "origin", "main"]:
             return SimpleNamespace(stdout="Updating\n", stderr="", returncode=0)
-        return SimpleNamespace(returncode=0)
+        return SimpleNamespace(returncode=0, stdout="", stderr="")
 
     monkeypatch.setattr(hermes_main.subprocess, "run", fake_run)
 

tests/hermes_cli/test_update_stale_dashboard.py

@@ -1,19 +1,62 @@
-"""Tests for _warn_stale_dashboard_processes — stale dashboard detection.
+"""Tests for the stale-dashboard handling run at the end of ``hermes update``.
 
-Ensures ``hermes update`` warns the user when dashboard processes from a
-previous version are still running after files on disk have been replaced.
-See #16872.
+``hermes update`` detects ``hermes dashboard`` processes left over from the
+previous version and kills them (SIGTERM + SIGKILL grace, or ``taskkill /F``
+on Windows).  Without this, the running backend silently serves stale Python
+against a freshly-updated JS bundle, producing 401s / empty data.
+
+History:
+- #16872 introduced the warn-only helper (``_warn_stale_dashboard_processes``).
+- #17049 fixed a Windows wmic UnicodeDecodeError crash on non-UTF-8 locales.
+- This file now also covers the kill semantics that replaced the warning.
 """
 
 from __future__ import annotations
 
+import importlib
 import os
 import sys
-from unittest.mock import patch, MagicMock
+from unittest.mock import patch, MagicMock, call
 
 import pytest
 
-from hermes_cli.main import _warn_stale_dashboard_processes
+from hermes_cli.main import (
+    _find_stale_dashboard_pids,
+    _kill_stale_dashboard_processes,
+    _warn_stale_dashboard_processes,  # back-compat alias
+)
+
+
+@pytest.fixture(autouse=True)
+def _refresh_bindings_against_live_module():
+    """Rebind module-level names to the *current* ``hermes_cli.main``.
+
+    Other tests in the suite (notably ``test_env_loader.py`` and
+    ``test_skills_subparser.py``) reload or delete ``hermes_cli.main`` from
+    ``sys.modules``.  When that happens on the same xdist worker before we
+    run, our top-of-file ``from hermes_cli.main import ...`` bindings end
+    up pointing at the *old* module object.  ``patch(\"hermes_cli.main.X\")``
+    then patches the *new* module, but the function we call still resolves
+    ``_find_stale_dashboard_pids`` via its stale ``__globals__``, so every
+    patch becomes a no-op and the kill path silently returns early.
+
+    Refreshing the bindings (and the patch target) to the live module
+    object — and keeping them consistent — makes the tests immune to
+    ordering within the worker.  The fix lives in the test module because
+    the two pollutants above are load-bearing for their own tests.
+    """
+    global _find_stale_dashboard_pids
+    global _kill_stale_dashboard_processes
+    global _warn_stale_dashboard_processes
+
+    live = sys.modules.get("hermes_cli.main")
+    if live is None:
+        live = importlib.import_module("hermes_cli.main")
+
+    _find_stale_dashboard_pids = live._find_stale_dashboard_pids
+    _kill_stale_dashboard_processes = live._kill_stale_dashboard_processes
+    _warn_stale_dashboard_processes = live._warn_stale_dashboard_processes
+    yield
 
 
 def _ps_line(pid: int, cmd: str) -> str:
@@ -21,11 +64,26 @@ def _ps_line(pid: int, cmd: str) -> str:
     return f"{pid:>7} {cmd}"
 
 
-class TestWarnStaleDashboardProcesses:
-    """Unit tests for the stale dashboard process warning."""
+def _ps_runner(stdout: str):
+    """Build a subprocess.run side_effect that only stubs ps -A calls.
 
-    def test_no_warning_when_no_dashboard_running(self, capsys):
-        """ps returns no matching processes — no warning should be printed."""
+    Any other subprocess.run invocation (e.g. taskkill on Windows) is
+    handed back as a successful no-op.  This lets tests exercise the real
+    scan path without having to re-stub every unrelated subprocess call
+    made later in ``_kill_stale_dashboard_processes``.
+    """
+    def _side_effect(args, *a, **kw):
+        if isinstance(args, (list, tuple)) and args and args[0] == "ps":
+            return MagicMock(returncode=0, stdout=stdout, stderr="")
+        # Any other subprocess.run (e.g. taskkill) — benign success stub.
+        return MagicMock(returncode=0, stdout="", stderr="")
+    return _side_effect
+
+
+class TestFindStaleDashboardPids:
+    """Unit tests for the ps/wmic-based detection step."""
+
+    def test_no_matches_returns_empty(self):
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
@@ -35,26 +93,18 @@ class TestWarnStaleDashboardProcesses:
                 + "\n",
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "dashboard process" not in output
+            assert _find_stale_dashboard_pids() == []
 
-    def test_warning_printed_for_running_dashboard(self, capsys):
-        """ps finds a dashboard PID — warning with PID should appear."""
+    def test_matches_running_dashboard(self):
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
                 stdout=_ps_line(12345, "python3 -m hermes_cli.main dashboard --port 9119") + "\n",
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "1 dashboard process" in output
-        assert "PID 12345" in output
-        assert "kill <pid>" in output
+            assert _find_stale_dashboard_pids() == [12345]
 
-    def test_multiple_dashboard_pids(self, capsys):
-        """Multiple dashboard processes — all PIDs listed."""
+    def test_multiple_matches(self):
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
@@ -65,15 +115,9 @@ class TestWarnStaleDashboardProcesses:
                 ]) + "\n",
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "3 dashboard process" in output
-        assert "PID 12345" in output
-        assert "PID 12346" in output
-        assert "PID 12347" in output
+            assert sorted(_find_stale_dashboard_pids()) == [12345, 12346, 12347]
 
-    def test_self_pid_excluded(self, capsys):
-        """The current process PID should not be reported."""
+    def test_self_pid_excluded(self):
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
@@ -83,41 +127,51 @@ class TestWarnStaleDashboardProcesses:
                 ]) + "\n",
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        # The self PID may still appear inside an unrelated context, so anchor
-        # the check to "PID <self>" which is how the warning prints.
-        assert f"PID {os.getpid()}" not in output
-        assert "PID 12345" in output
+            pids = _find_stale_dashboard_pids()
+        assert os.getpid() not in pids
+        assert 12345 in pids
 
-    def test_ps_not_found_silently_ignored(self, capsys):
-        """If ps is missing (FileNotFoundError), no crash, no warning."""
+    def test_ps_not_found_returns_empty(self):
         with patch("subprocess.run", side_effect=FileNotFoundError):
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert output == ""
+            assert _find_stale_dashboard_pids() == []
 
-    def test_ps_timeout_silently_ignored(self, capsys):
-        """If ps times out, no crash, no warning."""
+    def test_ps_timeout_returns_empty(self):
         import subprocess as sp
-
         with patch("subprocess.run", side_effect=sp.TimeoutExpired("ps", 10)):
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert output == ""
+            assert _find_stale_dashboard_pids() == []
 
-    def test_empty_ps_output_no_warning(self, capsys):
-        """ps returns 0 but empty stdout — no warning."""
+    def test_unrelated_process_containing_word_dashboard_not_matched(self):
+        """Guards against greedy pgrep-style matching catching chat sessions
+        or unrelated processes whose cmdline happens to contain 'dashboard'.
+        """
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
-                returncode=0, stdout="\n", stderr=""
+                returncode=0,
+                stdout="\n".join([
+                    _ps_line(12345, "python3 -m hermes_cli.main dashboard --port 9119"),
+                    _ps_line(22222, "python3 -m hermes_cli.main chat -q 'rewrite my dashboard'"),
+                    _ps_line(33333, "node /opt/grafana/dashboard-server.js"),
+                ]) + "\n",
+                stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "dashboard process" not in output
+            pids = _find_stale_dashboard_pids()
+        assert pids == [12345]
 
-    def test_invalid_pid_lines_skipped(self, capsys):
-        """Malformed ps lines should be skipped gracefully."""
+    def test_grep_lines_ignored(self):
+        with patch("subprocess.run") as mock_run:
+            mock_run.return_value = MagicMock(
+                returncode=0,
+                stdout="\n".join([
+                    _ps_line(99999, "grep hermes dashboard"),
+                    _ps_line(12345, "hermes dashboard --port 9119"),
+                ]) + "\n",
+                stderr="",
+            )
+            pids = _find_stale_dashboard_pids()
+        assert 99999 not in pids
+        assert 12345 in pids
+
+    def test_invalid_pid_lines_skipped(self):
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
@@ -128,50 +182,213 @@ class TestWarnStaleDashboardProcesses:
                 ]) + "\n",
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "PID 12345" in output
-        assert "1 dashboard process" in output
+            pids = _find_stale_dashboard_pids()
+        assert pids == [12345]
 
-    def test_unrelated_process_containing_word_dashboard_not_matched(self, capsys):
-        """A process whose cmdline contains 'dashboard' but isn't a hermes
-        dashboard process must NOT be flagged.  This guards against the old
-        ``pgrep -f "hermes.*dashboard"`` greedy regex that matched e.g. a
-        chat session argv containing both words.
-        """
+
+@pytest.mark.skipif(sys.platform == "win32", reason="POSIX kill semantics")
+class TestKillStaleDashboardPosix:
+    """Kill path on Linux / macOS: SIGTERM then SIGKILL any survivors."""
+
+    def test_no_stale_processes_is_a_noop(self, capsys):
+        with patch("hermes_cli.main._find_stale_dashboard_pids", return_value=[]):
+            _kill_stale_dashboard_processes()
+        assert capsys.readouterr().out == ""
+
+    def test_sigterm_graceful_exit(self, capsys):
+        """Processes that exit on SIGTERM (the probe gets ProcessLookupError)
+        are reported as stopped and SIGKILL is never sent."""
+        import signal as _signal
+
+        killed_signals: list[tuple[int, int]] = []
+
+        def fake_kill(pid, sig):
+            killed_signals.append((pid, sig))
+            if sig == 0:
+                # Probe after SIGTERM → "process gone".
+                raise ProcessLookupError
+            # SIGTERM itself: succeed silently.
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345, 12346]), \
+             patch("os.kill", side_effect=fake_kill), \
+             patch("time.sleep"):
+            _kill_stale_dashboard_processes()
+
+        # Both got SIGTERM.
+        sigterms = [pid for pid, sig in killed_signals if sig == _signal.SIGTERM]
+        assert sorted(sigterms) == [12345, 12346]
+        # No SIGKILL was needed.
+        assert not any(sig == _signal.SIGKILL for _, sig in killed_signals)
+
+        out = capsys.readouterr().out
+        assert "Stopping 2 dashboard" in out
+        assert "✓ stopped PID 12345" in out
+        assert "✓ stopped PID 12346" in out
+        assert "Restart the dashboard" in out
+
+    def test_sigkill_fallback_for_survivors(self, capsys):
+        """If a process survives SIGTERM + the grace window, SIGKILL is sent."""
+        import signal as _signal
+
+        sent: list[tuple[int, int]] = []
+
+        def fake_kill(pid, sig):
+            sent.append((pid, sig))
+            # Simulate stubborn process: probe (sig 0) always succeeds,
+            # SIGTERM does nothing, SIGKILL is where it "dies".
+            if sig in (_signal.SIGTERM, 0, _signal.SIGKILL):
+                return
+            # Any other signal — also fine.
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[99999]), \
+             patch("os.kill", side_effect=fake_kill), \
+             patch("time.sleep"), \
+             patch("time.monotonic", side_effect=[0.0] + [10.0] * 20):
+            # monotonic jumps past the 3s deadline on the second read so the
+            # grace loop exits immediately after one iteration.
+            _kill_stale_dashboard_processes()
+
+        signals_sent = [sig for _, sig in sent]
+        assert _signal.SIGTERM in signals_sent
+        assert _signal.SIGKILL in signals_sent
+
+        out = capsys.readouterr().out
+        assert "✓ stopped PID 99999" in out
+
+    def test_permission_error_is_reported_not_raised(self, capsys):
+        """os.kill raising PermissionError (e.g. another user's process)
+        must not abort hermes update — it's reported as a failure and we
+        move on."""
+        def fake_kill(pid, sig):
+            raise PermissionError("Operation not permitted")
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345]), \
+             patch("os.kill", side_effect=fake_kill), \
+             patch("time.sleep"):
+            _kill_stale_dashboard_processes()  # must not raise
+
+        out = capsys.readouterr().out
+        assert "✗ failed to stop PID 12345" in out
+        assert "Operation not permitted" in out
+
+    def test_process_already_gone_counts_as_stopped(self, capsys):
+        """ProcessLookupError on the initial SIGTERM means the process
+        already exited between detection and the kill — treat as success."""
+        def fake_kill(pid, sig):
+            raise ProcessLookupError
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345]), \
+             patch("os.kill", side_effect=fake_kill), \
+             patch("time.sleep"):
+            _kill_stale_dashboard_processes()
+
+        out = capsys.readouterr().out
+        assert "✓ stopped PID 12345" in out
+        assert "failed to stop" not in out
+
+
+class TestKillStaleDashboardWindows:
+    """Kill path on Windows: taskkill /F."""
+
+    def test_taskkill_invoked_for_each_pid(self, monkeypatch, capsys):
+        monkeypatch.setattr(sys, "platform", "win32")
+
+        def fake_run(args, *a, **kw):
+            # taskkill returns 0 on success
+            return MagicMock(returncode=0, stdout="", stderr="")
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345, 12346]), \
+             patch("subprocess.run", side_effect=fake_run) as mock_run:
+            _kill_stale_dashboard_processes()
+
+        # Each PID triggered a taskkill /PID <n> /F invocation.
+        taskkill_calls = [
+            c for c in mock_run.call_args_list
+            if c.args and isinstance(c.args[0], list) and c.args[0][:1] == ["taskkill"]
+        ]
+        assert len(taskkill_calls) == 2
+        assert ["taskkill", "/PID", "12345", "/F"] in [c.args[0] for c in taskkill_calls]
+        assert ["taskkill", "/PID", "12346", "/F"] in [c.args[0] for c in taskkill_calls]
+
+        out = capsys.readouterr().out
+        assert "✓ stopped PID 12345" in out
+        assert "✓ stopped PID 12346" in out
+
+    def test_taskkill_failure_is_reported(self, monkeypatch, capsys):
+        monkeypatch.setattr(sys, "platform", "win32")
+
+        def fake_run(args, *a, **kw):
+            return MagicMock(returncode=128, stdout="",
+                             stderr="ERROR: Access is denied.")
+
+        with patch("hermes_cli.main._find_stale_dashboard_pids",
+                   return_value=[12345]), \
+             patch("subprocess.run", side_effect=fake_run):
+            _kill_stale_dashboard_processes()  # must not raise
+
+        out = capsys.readouterr().out
+        assert "✗ failed to stop PID 12345" in out
+        assert "Access is denied" in out
+
+
+class TestBackCompatAlias:
+    """``_warn_stale_dashboard_processes`` is kept as an alias for the
+    new kill function so old imports don't break."""
+
+    def test_alias_is_the_kill_function(self):
+        assert _warn_stale_dashboard_processes is _kill_stale_dashboard_processes
+
+
+class TestWindowsWmicEncoding:
+    """Regression tests for #17049 — the Windows wmic branch must not crash
+    `hermes update` on non-UTF-8 system locales (e.g. cp936 on zh-CN).
+    """
+
+    def test_wmic_invoked_with_utf8_ignore_errors(self, monkeypatch):
+        """The wmic subprocess.run call must pass encoding='utf-8' and
+        errors='ignore' so the subprocess reader thread cannot raise
+        UnicodeDecodeError on non-UTF-8 wmic output."""
+        monkeypatch.setattr(sys, "platform", "win32")
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
                 returncode=0,
-                stdout="\n".join([
-                    # Legitimate dashboard — should match.
-                    _ps_line(12345, "python3 -m hermes_cli.main dashboard --port 9119"),
-                    # hermes running something else, with "dashboard" as a
-                    # substring of an unrelated arg — should NOT match.
-                    _ps_line(22222, "python3 -m hermes_cli.main chat -q 'rewrite my dashboard'"),
-                    # Completely unrelated process mentioning dashboard.
-                    _ps_line(33333, "node /opt/grafana/dashboard-server.js"),
-                ]) + "\n",
+                stdout=(
+                    "CommandLine=python -m hermes_cli.main dashboard\n"
+                    "ProcessId=12345\n"
+                ),
                 stderr="",
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "1 dashboard process" in output
-        assert "PID 12345" in output
-        assert "PID 22222" not in output
-        assert "PID 33333" not in output
+            _find_stale_dashboard_pids()
 
-    def test_grep_lines_ignored(self, capsys):
-        """Lines containing 'grep' (from a pipe in ps output) are ignored."""
+        # The wmic call is the first subprocess.run invocation.
+        assert mock_run.called, "subprocess.run was not invoked"
+        wmic_call = mock_run.call_args_list[0]
+        kwargs = wmic_call.kwargs
+        assert kwargs.get("encoding") == "utf-8", (
+            "encoding kwarg must be 'utf-8' so wmic output is decoded "
+            "deterministically rather than via the implicit reader-thread "
+            "default that crashes on non-UTF-8 locales (#17049)."
+        )
+        assert kwargs.get("errors") == "ignore", (
+            "errors kwarg must be 'ignore' so undecodable bytes don't take "
+            "down the reader thread (#17049)."
+        )
+
+    def test_wmic_returns_none_stdout_does_not_crash(self, monkeypatch):
+        """If subprocess.run returns successfully but stdout is None — which
+        is what Python 3.11 leaves behind when the reader thread silently
+        crashed on UnicodeDecodeError before this fix landed — detection
+        must short-circuit instead of raising AttributeError on
+        ``None.split('\\n')`` and aborting `hermes update` (#17049)."""
+        monkeypatch.setattr(sys, "platform", "win32")
         with patch("subprocess.run") as mock_run:
             mock_run.return_value = MagicMock(
-                returncode=0,
-                stdout="\n".join([
-                    _ps_line(99999, "grep hermes dashboard"),
-                    _ps_line(12345, "hermes dashboard --port 9119"),
-                ]) + "\n",
-                stderr="",
+                returncode=0, stdout=None, stderr=""
             )
-            _warn_stale_dashboard_processes()
-        output = capsys.readouterr().out
-        assert "PID 99999" not in output
-        assert "PID 12345" in output
+            # Must not raise.
+            assert _find_stale_dashboard_pids() == []

tests/hermes_cli/test_user_providers_model_switch.py

@@ -453,6 +453,142 @@ def test_list_authenticated_providers_no_duplicate_labels_across_schemas(monkeyp
     )
 
 
+def test_list_authenticated_providers_hides_custom_shadowing_builtin_endpoint(monkeypatch):
+    """#16970: a custom_providers entry whose ``base_url`` matches a built-in
+    provider's endpoint should be hidden. The built-in row already represents
+    that endpoint with its canonical slug, curated model list, and auth wiring.
+
+    Repro: user sets ``DASHSCOPE_API_KEY`` (triggers the built-in ``alibaba``
+    row pointing at the static ``inference_base_url``) AND defines a
+    ``my-alibaba`` custom provider pointing at the same URL. Before the fix,
+    the picker showed both rows for one endpoint.
+    """
+    monkeypatch.setenv("DASHSCOPE_API_KEY", "sk-test")
+    monkeypatch.setattr(
+        "agent.models_dev.fetch_models_dev",
+        lambda: {
+            "alibaba": {
+                "name": "Alibaba Cloud (DashScope)",
+                "env": ["DASHSCOPE_API_KEY"],
+            }
+        },
+    )
+    monkeypatch.setattr("hermes_cli.providers.HERMES_OVERLAYS", {})
+
+    custom_providers = [
+        {
+            "name": "my-alibaba",
+            # Matches PROVIDER_REGISTRY['alibaba'].inference_base_url exactly.
+            "base_url": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
+            "api_key": "sk-sp-test",
+            "model": "qwen3.6-plus",
+            "models": {"qwen3.6-plus": {"context_length": 500000}},
+        }
+    ]
+
+    providers = list_authenticated_providers(
+        current_provider="my-alibaba",
+        user_providers={},
+        custom_providers=custom_providers,
+        max_models=50,
+    )
+
+    slugs = [p["slug"] for p in providers]
+    # Built-in alibaba row should be present.
+    assert "alibaba" in slugs, (
+        f"Expected built-in alibaba row, got slugs: {slugs}"
+    )
+    # Custom shadow row should be hidden — its base_url matches the built-in's.
+    assert not any("my-alibaba" in s for s in slugs), (
+        f"Custom my-alibaba should have been dedup'd against the built-in "
+        f"alibaba endpoint, got slugs: {slugs}"
+    )
+
+
+def test_list_authenticated_providers_keeps_custom_with_distinct_endpoint(monkeypatch):
+    """Dedup must only apply when the endpoint matches a built-in. A custom
+    provider on a genuinely distinct endpoint stays visible even if a
+    built-in is also authenticated."""
+    monkeypatch.setenv("DASHSCOPE_API_KEY", "sk-test")
+    monkeypatch.setattr(
+        "agent.models_dev.fetch_models_dev",
+        lambda: {
+            "alibaba": {
+                "name": "Alibaba Cloud (DashScope)",
+                "env": ["DASHSCOPE_API_KEY"],
+            }
+        },
+    )
+    monkeypatch.setattr("hermes_cli.providers.HERMES_OVERLAYS", {})
+
+    custom_providers = [
+        {
+            "name": "my-private-relay",
+            "base_url": "https://relay.example.internal/v1",
+            "api_key": "sk-relay-test",
+            "model": "qwen3.6-plus",
+            "models": {"qwen3.6-plus": {}},
+        }
+    ]
+
+    providers = list_authenticated_providers(
+        current_provider="my-private-relay",
+        user_providers={},
+        custom_providers=custom_providers,
+        max_models=50,
+    )
+
+    slugs = [p["slug"] for p in providers]
+    assert any("my-private-relay" in s for s in slugs), (
+        f"Custom provider on distinct endpoint must stay visible, got: {slugs}"
+    )
+
+
+def test_list_authenticated_providers_dedup_honors_base_url_env_override(monkeypatch):
+    """The dedup must track the EFFECTIVE endpoint — if DASHSCOPE_BASE_URL
+    overrides the static inference_base_url, a custom provider pointing at
+    the overridden URL (not the static one) should still be recognized as
+    a duplicate."""
+    monkeypatch.setenv("DASHSCOPE_API_KEY", "sk-test")
+    monkeypatch.setenv(
+        "DASHSCOPE_BASE_URL",
+        "https://custom-dashscope.example.com/v1",
+    )
+    monkeypatch.setattr(
+        "agent.models_dev.fetch_models_dev",
+        lambda: {
+            "alibaba": {
+                "name": "Alibaba Cloud (DashScope)",
+                "env": ["DASHSCOPE_API_KEY"],
+            }
+        },
+    )
+    monkeypatch.setattr("hermes_cli.providers.HERMES_OVERLAYS", {})
+
+    custom_providers = [
+        {
+            "name": "my-dashscope-override",
+            # Same URL as DASHSCOPE_BASE_URL env override above.
+            "base_url": "https://custom-dashscope.example.com/v1",
+            "api_key": "sk-test",
+            "model": "qwen3.6-plus",
+        }
+    ]
+
+    providers = list_authenticated_providers(
+        current_provider="alibaba",
+        user_providers={},
+        custom_providers=custom_providers,
+        max_models=50,
+    )
+
+    slugs = [p["slug"] for p in providers]
+    assert not any("my-dashscope-override" in s for s in slugs), (
+        f"Custom entry matching env-overridden built-in endpoint should be "
+        f"dedup'd, got: {slugs}"
+    )
+
+
 # =============================================================================
 # Tests for _get_named_custom_provider with providers: dict
 # =============================================================================

tests/hermes_cli/test_web_server.py

@@ -29,7 +29,7 @@ class TestReloadEnv:
         """reload_env() adds vars from .env that are not in os.environ."""
         env_file = tmp_path / ".env"
         env_file.write_text("TEST_RELOAD_VAR=hello123\n")
-        with patch("hermes_cli.config.get_env_path", return_value=env_file):
+        with patch.dict(reload_env.__globals__, {"get_env_path": lambda: env_file}):
             os.environ.pop("TEST_RELOAD_VAR", None)
             count = reload_env()
             assert count >= 1
@@ -40,7 +40,7 @@ class TestReloadEnv:
         """reload_env() updates vars whose value changed on disk."""
         env_file = tmp_path / ".env"
         env_file.write_text("TEST_RELOAD_VAR=old_value\n")
-        with patch("hermes_cli.config.get_env_path", return_value=env_file):
+        with patch.dict(reload_env.__globals__, {"get_env_path": lambda: env_file}):
             os.environ["TEST_RELOAD_VAR"] = "old_value"
             # Now change the file
             env_file.write_text("TEST_RELOAD_VAR=new_value\n")
@@ -55,7 +55,7 @@ class TestReloadEnv:
         env_file.write_text("")  # empty .env
         # Pick a known key from OPTIONAL_ENV_VARS
         known_key = next(iter(OPTIONAL_ENV_VARS.keys()))
-        with patch("hermes_cli.config.get_env_path", return_value=env_file):
+        with patch.dict(reload_env.__globals__, {"get_env_path": lambda: env_file}):
             os.environ[known_key] = "stale_value"
             count = reload_env()
             assert known_key not in os.environ
@@ -65,7 +65,7 @@ class TestReloadEnv:
         """reload_env() preserves non-Hermes env vars even when absent from .env."""
         env_file = tmp_path / ".env"
         env_file.write_text("")
-        with patch("hermes_cli.config.get_env_path", return_value=env_file):
+        with patch.dict(reload_env.__globals__, {"get_env_path": lambda: env_file}):
             os.environ["MY_CUSTOM_UNRELATED_VAR"] = "keep_me"
             reload_env()
             assert os.environ.get("MY_CUSTOM_UNRELATED_VAR") == "keep_me"
@@ -371,6 +371,12 @@ class TestBuildSchemaFromConfig:
             assert entry["type"] == "select"
             assert "options" in entry
             assert "local" in entry["options"]
+            assert "vercel_sandbox" in entry["options"]
+        runtime_entry = CONFIG_SCHEMA["terminal.vercel_runtime"]
+        assert runtime_entry["type"] == "select"
+        assert "node24" in runtime_entry["options"]
+        assert "python3.13" in runtime_entry["options"]
+        assert len(runtime_entry["options"]) >= 3
 
     def test_empty_prefix_produces_correct_keys(self):
         from hermes_cli.web_server import _build_schema_from_config
@@ -671,8 +677,12 @@ class TestNewEndpoints:
         assert resp.status_code == 200
         assert resp.json()["command"] == "hermes setup"
 
-    def test_profiles_create_creates_wrapper_alias_when_safe(self):
-        from pathlib import Path
+    def test_profiles_create_creates_wrapper_alias_when_safe(self, monkeypatch, tmp_path):
+        import hermes_cli.profiles as profiles_mod
+
+        wrapper_dir = tmp_path / "bin"
+        wrapper_dir.mkdir()
+        monkeypatch.setattr(profiles_mod, "_get_wrapper_dir", lambda: wrapper_dir)
 
         resp = self.client.post(
             "/api/profiles",
@@ -680,7 +690,7 @@ class TestNewEndpoints:
         )
 
         assert resp.status_code == 200
-        wrapper_path = Path.home() / ".local" / "bin" / "writer"
+        wrapper_path = wrapper_dir / "writer"
         assert wrapper_path.exists()
         assert wrapper_path.read_text() == '#!/bin/sh\nexec hermes -p writer "$@"\n'
 
@@ -2057,14 +2067,24 @@ class TestPtyWebSocket:
             assert b"round-trip-payload" in buf
 
     def test_resize_escape_is_forwarded(self, monkeypatch):
-        # Resize escape gets intercepted and applied via TIOCSWINSZ,
-        # then ``tput cols/lines`` reports the new dimensions back.
+        # Resize escape gets intercepted and applied via TIOCSWINSZ, then the
+        # child reads the TTY ioctl directly. Avoid tput because CI may not set
+        # TERM for non-interactive shells.
+        import sys
+
+        winsize_script = (
+            "import fcntl, struct, termios, time; "
+            "time.sleep(0.15); "
+            "rows, cols, *_ = struct.unpack('HHHH', "
+            "fcntl.ioctl(0, termios.TIOCGWINSZ, b'\\0' * 8)); "
+            "print(cols); print(rows)"
+        )
         monkeypatch.setattr(
             self.ws_module,
             "_resolve_chat_argv",
-            # sleep gives the test time to push the resize before tput runs
+            # sleep gives the test time to push the resize before the child reads the ioctl.
             lambda resume=None, sidecar_url=None: (
-                ["/bin/sh", "-c", "sleep 0.15; tput cols; tput lines"],
+                [sys.executable, "-c", winsize_script],
                 None,
                 None,
             ),
@@ -2153,13 +2173,30 @@ class TestPtyWebSocket:
     def test_pub_broadcasts_to_events_subscribers(self, monkeypatch):
         """Frame written to /api/pub is rebroadcast verbatim to every
         /api/events subscriber on the same channel."""
+        import time
         from urllib.parse import urlencode
+        from hermes_cli import web_server as ws_mod
 
         qs = urlencode({"token": self.token, "channel": "broadcast-test"})
         pub_path = f"/api/pub?{qs}"
         sub_path = f"/api/events?{qs}"
 
         with self.client.websocket_connect(sub_path) as sub:
+            # Wait for the subscriber to be registered on the server side.
+            # websocket_connect returns when ws.accept() completes, but the
+            # server adds us to ``_event_channels`` in a follow-up await,
+            # so a publish immediately after connect can race ahead of the
+            # subscriber registration and the message is dropped.
+            deadline = time.monotonic() + 5.0
+            while time.monotonic() < deadline:
+                if ws_mod._event_channels.get("broadcast-test"):
+                    break
+                time.sleep(0.01)
+            else:
+                raise AssertionError(
+                    "subscriber did not register on channel within 5s"
+                )
+
             with self.client.websocket_connect(pub_path) as pub:
                 pub.send_text('{"type":"tool.start","payload":{"tool_id":"t1"}}')
                 received = sub.receive_text()