fix(ui-tui): harden TUI error handling, model validation, command UX parity, and gateway lifecycle

cli.py

@@ -1194,6 +1194,10 @@ def _resolve_attachment_path(raw_path: str) -> Path | None:
         return None
 
     expanded = os.path.expandvars(os.path.expanduser(token))
+    if os.name != "nt":
+        normalized = expanded.replace("\\", "/")
+        if len(normalized) >= 3 and normalized[1] == ":" and normalized[2] == "/" and normalized[0].isalpha():
+            expanded = f"/mnt/{normalized[0].lower()}/{normalized[3:]}"
     path = Path(expanded)
     if not path.is_absolute():
         base_dir = Path(os.getenv("TERMINAL_CWD", os.getcwd()))
@@ -1276,10 +1280,12 @@ def _detect_file_drop(user_input: str) -> "dict | None":
         or stripped.startswith("~")
         or stripped.startswith("./")
         or stripped.startswith("../")
+        or (len(stripped) >= 3 and stripped[1] == ":" and stripped[2] in ("\\", "/") and stripped[0].isalpha())
         or stripped.startswith('"/')
         or stripped.startswith('"~')
         or stripped.startswith("'/")
         or stripped.startswith("'~")
+        or (len(stripped) >= 4 and stripped[0] in ("'", '"') and stripped[2] == ":" and stripped[3] in ("\\", "/") and stripped[1].isalpha())
     )
     if not starts_like_path:
         return None

hermes_cli/commands.py

@@ -73,7 +73,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
                args_hint="[focus topic]"),
     CommandDef("rollback", "List or restore filesystem checkpoints", "Session",
                args_hint="[number]"),
-    CommandDef("stop", "Kill all running background processes", "Session"),
+    CommandDef("stop", "Kill all running registered subprocesses", "Session"),
     CommandDef("approve", "Approve a pending dangerous command", "Session",
                gateway_only=True, args_hint="[session|always]"),
     CommandDef("deny", "Deny a pending dangerous command", "Session",
@@ -96,7 +96,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
     # Configuration
     CommandDef("config", "Show current configuration", "Configuration",
                cli_only=True),
-    CommandDef("model", "Switch model for this session", "Configuration", args_hint="[model] [--global]"),
+    CommandDef("model", "Switch model for this session", "Configuration", args_hint="[model] [--provider name] [--global]"),
     CommandDef("provider", "Show available providers and current provider",
                "Configuration"),
 
@@ -152,7 +152,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
                cli_only=True, aliases=("gateway",)),
     CommandDef("copy", "Copy the last assistant response to clipboard", "Info",
                cli_only=True, args_hint="[number]"),
-    CommandDef("paste", "Check clipboard for an image and attach it", "Info",
+    CommandDef("paste", "Attach clipboard image or manage text paste shelf", "Info",
                cli_only=True),
     CommandDef("image", "Attach a local image file for your next prompt", "Info",
                cli_only=True, args_hint="<path>"),

hermes_cli/models.py

@@ -1803,8 +1803,8 @@ def validate_requested_model(
                 )
 
             return {
-                "accepted": True,
-                "persist": True,
+                "accepted": False,
+                "persist": False,
                 "recognized": False,
                 "message": message,
             }
@@ -1817,8 +1817,8 @@ def validate_requested_model(
             message += f"\n  If this server expects `/v1`, try base URL: `{probe.get('suggested_base_url')}`"
 
         return {
-            "accepted": True,
-            "persist": True,
+            "accepted": False,
+            "persist": False,
             "recognized": False,
             "message": message,
         }
@@ -1882,8 +1882,8 @@ def validate_requested_model(
     # but warn so typos don't silently break things.
     provider_label = _PROVIDER_LABELS.get(normalized, normalized)
     return {
-        "accepted": True,
-        "persist": True,
+        "accepted": False,
+        "persist": False,
         "recognized": False,
         "message": (
             f"Could not reach the {provider_label} API to validate `{requested}`. "

tests/hermes_cli/test_model_validation.py

@@ -437,33 +437,33 @@ class TestValidateApiNotFound:
     def test_warning_includes_suggestions(self):
         result = _validate("anthropic/claude-opus-4.5")
         assert result["accepted"] is False
+        assert result["persist"] is False
         assert "Similar models" in result["message"]
 
 
-# -- validate — API unreachable — accept and persist everything ----------------
+# -- validate — API unreachable — reject with guidance ----------------
 
 class TestValidateApiFallback:
-    def test_any_model_accepted_when_api_down(self):
+    def test_any_model_rejected_when_api_down(self):
         result = _validate("anthropic/claude-opus-4.6", api_models=None)
-        assert result["accepted"] is True
-        assert result["persist"] is True
+        assert result["accepted"] is False
+        assert result["persist"] is False
 
-    def test_unknown_model_also_accepted_when_api_down(self):
-        """No hardcoded catalog gatekeeping — accept, persist, and warn."""
+    def test_unknown_model_also_rejected_when_api_down(self):
         result = _validate("anthropic/claude-next-gen", api_models=None)
-        assert result["accepted"] is True
-        assert result["persist"] is True
+        assert result["accepted"] is False
+        assert result["persist"] is False
         assert "could not reach" in result["message"].lower()
 
-    def test_zai_model_accepted_when_api_down(self):
+    def test_zai_model_rejected_when_api_down(self):
         result = _validate("glm-5", provider="zai", api_models=None)
-        assert result["accepted"] is True
-        assert result["persist"] is True
+        assert result["accepted"] is False
+        assert result["persist"] is False
 
-    def test_unknown_provider_accepted_when_api_down(self):
+    def test_unknown_provider_rejected_when_api_down(self):
         result = _validate("some-model", provider="totally-unknown", api_models=None)
-        assert result["accepted"] is True
-        assert result["persist"] is True
+        assert result["accepted"] is False
+        assert result["persist"] is False
 
     def test_custom_endpoint_warns_with_probed_url_and_v1_hint(self):
         with patch(
@@ -483,7 +483,7 @@ class TestValidateApiFallback:
                 base_url="http://localhost:8000",
             )
 
-        assert result["accepted"] is True
-        assert result["persist"] is True
+        assert result["accepted"] is False
+        assert result["persist"] is False
         assert "http://localhost:8000/v1/models" in result["message"]
         assert "http://localhost:8000/v1" in result["message"]

tests/test_tui_gateway_server.py

@@ -167,11 +167,87 @@ def test_config_set_model_uses_live_switch_path(monkeypatch):
     assert seen["args"] == ("sid", "session-key", "new/model")
 
 
+def test_config_set_model_global_persists(monkeypatch):
+    class _Agent:
+        provider = "openrouter"
+        model = "old/model"
+        base_url = ""
+        api_key = "sk-old"
+
+        def switch_model(self, **kwargs):
+            return None
+
+    result = types.SimpleNamespace(
+        success=True,
+        new_model="anthropic/claude-sonnet-4.6",
+        target_provider="anthropic",
+        api_key="sk-new",
+        base_url="https://api.anthropic.com",
+        api_mode="anthropic_messages",
+        warning_message="",
+    )
+    seen = {}
+    saved = {}
+
+    def _switch_model(**kwargs):
+        seen.update(kwargs)
+        return result
+
+    server._sessions["sid"] = _session(agent=_Agent())
+    monkeypatch.setattr("hermes_cli.model_switch.switch_model", _switch_model)
+    monkeypatch.setattr(server, "_restart_slash_worker", lambda session: None)
+    monkeypatch.setattr(server, "_emit", lambda *args, **kwargs: None)
+    monkeypatch.setattr("hermes_cli.config.save_config", lambda cfg: saved.update(cfg))
+
+    resp = server.handle_request(
+        {"id": "1", "method": "config.set", "params": {"session_id": "sid", "key": "model", "value": "anthropic/claude-sonnet-4.6 --global"}}
+    )
+
+    assert resp["result"]["value"] == "anthropic/claude-sonnet-4.6"
+    assert seen["is_global"] is True
+    assert saved["model"]["default"] == "anthropic/claude-sonnet-4.6"
+    assert saved["model"]["provider"] == "anthropic"
+    assert saved["model"]["base_url"] == "https://api.anthropic.com"
+
+
+def test_config_set_personality_rejects_unknown_name(monkeypatch):
+    monkeypatch.setattr(server, "_available_personalities", lambda cfg=None: {"helpful": "You are helpful."})
+    resp = server.handle_request(
+        {"id": "1", "method": "config.set", "params": {"key": "personality", "value": "bogus"}}
+    )
+
+    assert "error" in resp
+    assert "Unknown personality" in resp["error"]["message"]
+
+
+def test_config_set_personality_resets_history_and_returns_info(monkeypatch):
+    session = _session(agent=types.SimpleNamespace(), history=[{"role": "user", "text": "hi"}], history_version=4)
+    new_agent = types.SimpleNamespace(model="x")
+    emits = []
+
+    server._sessions["sid"] = session
+    monkeypatch.setattr(server, "_available_personalities", lambda cfg=None: {"helpful": "You are helpful."})
+    monkeypatch.setattr(server, "_make_agent", lambda sid, key, session_id=None: new_agent)
+    monkeypatch.setattr(server, "_session_info", lambda agent: {"model": getattr(agent, "model", "?")})
+    monkeypatch.setattr(server, "_restart_slash_worker", lambda session: None)
+    monkeypatch.setattr(server, "_emit", lambda *args: emits.append(args))
+
+    resp = server.handle_request(
+        {"id": "1", "method": "config.set", "params": {"session_id": "sid", "key": "personality", "value": "helpful"}}
+    )
+
+    assert resp["result"]["history_reset"] is True
+    assert resp["result"]["info"] == {"model": "x"}
+    assert session["history"] == []
+    assert session["history_version"] == 5
+    assert ("session.info", "sid", {"model": "x"}) in emits
+
+
 def test_session_compress_uses_compress_helper(monkeypatch):
     agent = types.SimpleNamespace()
     server._sessions["sid"] = _session(agent=agent)
 
-    monkeypatch.setattr(server, "_compress_session_history", lambda session: (2, {"total": 42}))
+    monkeypatch.setattr(server, "_compress_session_history", lambda session, focus_topic=None: (2, {"total": 42}))
     monkeypatch.setattr(server, "_session_info", lambda _agent: {"model": "x"})
 
     with patch("tui_gateway.server._emit") as emit:
@@ -266,6 +342,36 @@ def test_image_attach_appends_local_image(monkeypatch):
     assert len(server._sessions["sid"]["attached_images"]) == 1
 
 
+def test_command_dispatch_exec_nonzero_surfaces_error(monkeypatch):
+    monkeypatch.setattr(server, "_load_cfg", lambda: {"quick_commands": {"boom": {"type": "exec", "command": "boom"}}})
+    monkeypatch.setattr(
+        server.subprocess,
+        "run",
+        lambda *args, **kwargs: types.SimpleNamespace(returncode=1, stdout="", stderr="failed"),
+    )
+
+    resp = server.handle_request({"id": "1", "method": "command.dispatch", "params": {"name": "boom"}})
+
+    assert "error" in resp
+    assert "failed" in resp["error"]["message"]
+
+
+def test_plugins_list_surfaces_loader_error(monkeypatch):
+    with patch("hermes_cli.plugins.get_plugin_manager", side_effect=Exception("boom")):
+        resp = server.handle_request({"id": "1", "method": "plugins.list", "params": {}})
+
+    assert "error" in resp
+    assert "boom" in resp["error"]["message"]
+
+
+def test_complete_slash_surfaces_completer_error(monkeypatch):
+    with patch("hermes_cli.commands.SlashCommandCompleter", side_effect=Exception("no completer")):
+        resp = server.handle_request({"id": "1", "method": "complete.slash", "params": {"text": "/mo"}})
+
+    assert "error" in resp
+    assert "no completer" in resp["error"]["message"]
+
+
 def test_input_detect_drop_attaches_image(monkeypatch):
     fake_cli = types.ModuleType("cli")
     fake_cli._detect_file_drop = lambda raw: {

tui_gateway/server.py

@@ -183,10 +183,19 @@ def handle_request(req: dict) -> dict | None:
 
 
 def _sess(params, rid):
-    s = _sessions.get(params.get("session_id", ""))
+    s = _sessions.get(params.get("session_id") or "")
     return (s, None) if s else (None, _err(rid, 4001, "session not found"))
 
 
+def _normalize_completion_path(path_part: str) -> str:
+    expanded = os.path.expanduser(path_part)
+    if os.name != "nt":
+        normalized = expanded.replace("\\", "/")
+        if len(normalized) >= 3 and normalized[1] == ":" and normalized[2] == "/" and normalized[0].isalpha():
+            return f"/mnt/{normalized[0].lower()}/{normalized[3:]}"
+    return expanded
+
+
 # ── Config I/O ────────────────────────────────────────────────────────
 
 def _load_cfg() -> dict:
@@ -327,38 +336,75 @@ def _restart_slash_worker(session: dict):
         session["slash_worker"] = None
 
 
-def _apply_model_switch(sid: str, session: dict, raw_input: str) -> dict:
-    agent = session.get("agent")
-    if not agent:
-        os.environ["HERMES_MODEL"] = raw_input
-        return {"value": raw_input, "warning": ""}
+def _persist_model_switch(result) -> None:
+    from hermes_cli.config import save_config
 
-    from hermes_cli.model_switch import switch_model
+    cfg = _load_cfg()
+    model_cfg = cfg.get("model")
+    if not isinstance(model_cfg, dict):
+        model_cfg = {}
+        cfg["model"] = model_cfg
+
+    model_cfg["default"] = result.new_model
+    model_cfg["provider"] = result.target_provider
+    if result.base_url:
+        model_cfg["base_url"] = result.base_url
+    else:
+        model_cfg.pop("base_url", None)
+    save_config(cfg)
+
+
+def _apply_model_switch(sid: str, session: dict, raw_input: str) -> dict:
+    from hermes_cli.model_switch import parse_model_flags, switch_model
+    from hermes_cli.runtime_provider import resolve_runtime_provider
+
+    model_input, explicit_provider, persist_global = parse_model_flags(raw_input)
+    if not model_input:
+        raise ValueError("model value required")
+
+    agent = session.get("agent")
+    if agent:
+        current_provider = getattr(agent, "provider", "") or ""
+        current_model = getattr(agent, "model", "") or ""
+        current_base_url = getattr(agent, "base_url", "") or ""
+        current_api_key = getattr(agent, "api_key", "") or ""
+    else:
+        runtime = resolve_runtime_provider(requested=None)
+        current_provider = str(runtime.get("provider", "") or "")
+        current_model = _resolve_model()
+        current_base_url = str(runtime.get("base_url", "") or "")
+        current_api_key = str(runtime.get("api_key", "") or "")
 
     result = switch_model(
-        raw_input=raw_input,
-        current_provider=getattr(agent, "provider", "") or "",
-        current_model=getattr(agent, "model", "") or "",
-        current_base_url=getattr(agent, "base_url", "") or "",
-        current_api_key=getattr(agent, "api_key", "") or "",
+        raw_input=model_input,
+        current_provider=current_provider,
+        current_model=current_model,
+        current_base_url=current_base_url,
+        current_api_key=current_api_key,
+        is_global=persist_global,
+        explicit_provider=explicit_provider,
     )
     if not result.success:
         raise ValueError(result.error_message or "model switch failed")
 
-    agent.switch_model(
-        new_model=result.new_model,
-        new_provider=result.target_provider,
-        api_key=result.api_key,
-        base_url=result.base_url,
-        api_mode=result.api_mode,
-    )
+    if agent:
+        agent.switch_model(
+            new_model=result.new_model,
+            new_provider=result.target_provider,
+            api_key=result.api_key,
+            base_url=result.base_url,
+            api_mode=result.api_mode,
+        )
+        _restart_slash_worker(session)
+        _emit("session.info", sid, _session_info(agent))
+
     os.environ["HERMES_MODEL"] = result.new_model
-    _restart_slash_worker(session)
-    _emit("session.info", sid, _session_info(agent))
+    if persist_global:
+        _persist_model_switch(result)
     return {"value": result.new_model, "warning": result.warning_message or ""}
 
 
-def _compress_session_history(session: dict) -> tuple[int, dict]:
+def _compress_session_history(session: dict, focus_topic: str | None = None) -> tuple[int, dict]:
     from agent.model_metadata import estimate_messages_tokens_rough
 
     agent = session["agent"]
@@ -370,6 +416,7 @@ def _compress_session_history(session: dict) -> tuple[int, dict]:
         history,
         getattr(agent, "_cached_system_prompt", "") or "",
         approx_tokens=approx_tokens,
+        focus_topic=focus_topic or None,
     )
     session["history"] = compressed
     session["history_version"] = int(session.get("history_version", 0)) + 1
@@ -617,21 +664,91 @@ def _resolve_personality_prompt(cfg: dict) -> str:
     if not name or name in ("default", "none", "neutral"):
         return ""
     try:
-        from hermes_cli.config import load_config as _load_full_cfg
-        personalities = _load_full_cfg().get("agent", {}).get("personalities", {})
+        from cli import load_cli_config
+
+        personalities = load_cli_config().get("agent", {}).get("personalities", {})
     except Exception:
-        personalities = cfg.get("agent", {}).get("personalities", {})
+        try:
+            from hermes_cli.config import load_config as _load_full_cfg
+
+            personalities = _load_full_cfg().get("agent", {}).get("personalities", {})
+        except Exception:
+            personalities = cfg.get("agent", {}).get("personalities", {})
     pval = personalities.get(name)
     if pval is None:
         return ""
-    if isinstance(pval, dict):
-        parts = [pval.get("system_prompt", "")]
-        if pval.get("tone"):
-            parts.append(f'Tone: {pval["tone"]}')
-        if pval.get("style"):
-            parts.append(f'Style: {pval["style"]}')
+    return _render_personality_prompt(pval)
+
+
+def _render_personality_prompt(value) -> str:
+    if isinstance(value, dict):
+        parts = [value.get("system_prompt", "")]
+        if value.get("tone"):
+            parts.append(f'Tone: {value["tone"]}')
+        if value.get("style"):
+            parts.append(f'Style: {value["style"]}')
         return "\n".join(p for p in parts if p)
-    return str(pval)
+    return str(value)
+
+
+def _available_personalities(cfg: dict | None = None) -> dict:
+    try:
+        from cli import load_cli_config
+
+        return load_cli_config().get("agent", {}).get("personalities", {}) or {}
+    except Exception:
+        try:
+            from hermes_cli.config import load_config as _load_full_cfg
+
+            return _load_full_cfg().get("agent", {}).get("personalities", {}) or {}
+        except Exception:
+            cfg = cfg or _load_cfg()
+            return cfg.get("agent", {}).get("personalities", {}) or {}
+
+
+def _validate_personality(value: str, cfg: dict | None = None) -> tuple[str, str]:
+    raw = str(value or "").strip()
+    name = raw.lower()
+    if not name or name in ("none", "default", "neutral"):
+        return "", ""
+
+    personalities = _available_personalities(cfg)
+    if name not in personalities:
+        names = sorted(personalities)
+        available = ", ".join(f"`{n}`" for n in names)
+        base = f"Unknown personality: `{raw}`."
+        if available:
+            base += f"\n\nAvailable: `none`, {available}"
+        else:
+            base += "\n\nNo personalities configured."
+        raise ValueError(base)
+
+    return name, _render_personality_prompt(personalities[name])
+
+
+def _apply_personality_to_session(sid: str, session: dict, new_prompt: str) -> tuple[bool, dict | None]:
+    if not session:
+        return False, None
+
+    try:
+        new_agent = _make_agent(sid, session["session_key"], session_id=session["session_key"])
+        session["agent"] = new_agent
+        with session["history_lock"]:
+            session["history"] = []
+            session["history_version"] = int(session.get("history_version", 0)) + 1
+        info = _session_info(new_agent)
+        _emit("session.info", sid, info)
+        _restart_slash_worker(session)
+        return True, info
+    except Exception:
+        if session.get("agent"):
+            agent = session["agent"]
+            agent.ephemeral_system_prompt = new_prompt or None
+            agent._cached_system_prompt = None
+            info = _session_info(agent)
+            _emit("session.info", sid, info)
+            return False, info
+        return False, None
 
 
 def _make_agent(sid: str, key: str, session_id: str | None = None):
@@ -893,9 +1010,11 @@ def _(rid, params: dict) -> dict:
         return err
     try:
         with session["history_lock"]:
-            removed, usage = _compress_session_history(session)
-        _emit("session.info", params.get("session_id", ""), _session_info(session["agent"]))
-        return _ok(rid, {"status": "compressed", "removed": removed, "usage": usage})
+            removed, usage = _compress_session_history(session, str(params.get("focus_topic", "") or "").strip())
+            messages = list(session.get("history", []))
+        info = _session_info(session["agent"])
+        _emit("session.info", params.get("session_id", ""), info)
+        return _ok(rid, {"status": "compressed", "removed": removed, "usage": usage, "info": info, "messages": messages})
     except Exception as e:
         return _err(rid, 5005, str(e))
 
@@ -906,7 +1025,7 @@ def _(rid, params: dict) -> dict:
     if err:
         return err
     import time as _time
-    filename = f"hermes_conversation_{_time.strftime('%Y%m%d_%H%M%S')}.json"
+    filename = os.path.abspath(f"hermes_conversation_{_time.strftime('%Y%m%d_%H%M%S')}.json")
     try:
         with open(filename, "w") as f:
             json.dump({"model": getattr(session["agent"], "model", ""), "messages": session.get("history", [])},
@@ -916,6 +1035,27 @@ def _(rid, params: dict) -> dict:
         return _err(rid, 5011, str(e))
 
 
+@method("session.close")
+def _(rid, params: dict) -> dict:
+    sid = params.get("session_id", "")
+    session = _sessions.pop(sid, None)
+    if not session:
+        return _ok(rid, {"closed": False})
+    try:
+        from tools.approval import unregister_gateway_notify
+
+        unregister_gateway_notify(session["session_key"])
+    except Exception:
+        pass
+    try:
+        worker = session.get("slash_worker")
+        if worker:
+            worker.close()
+    except Exception:
+        pass
+    return _ok(rid, {"closed": True})
+
+
 @method("session.branch")
 def _(rid, params: dict) -> dict:
     session, err = _sess(params, rid)
@@ -1087,6 +1227,7 @@ def _(rid, params: dict) -> dict:
 
     # Save-first: mirrors CLI keybinding path; more robust than has_image() precheck
     if not save_clipboard_image(img_path):
+        session["image_counter"] = max(0, session["image_counter"] - 1)
         msg = "Clipboard has image but extraction failed" if has_clipboard_image() else "No image found in clipboard"
         return _ok(rid, {"attached": False, "message": msg})
 
@@ -1182,6 +1323,9 @@ def _(rid, params: dict) -> dict:
 
 @method("prompt.background")
 def _(rid, params: dict) -> dict:
+    session, err = _sess(params, rid)
+    if err:
+        return err
     text, parent = params.get("text", ""), params.get("session_id", "")
     if not text:
         return _err(rid, 4012, "text required")
@@ -1275,8 +1419,7 @@ def _(rid, params: dict) -> dict:
             if session:
                 result = _apply_model_switch(params.get("session_id", ""), session, value)
             else:
-                os.environ["HERMES_MODEL"] = value
-                result = {"value": value, "warning": ""}
+                result = _apply_model_switch("", {"agent": None}, value)
             return _ok(rid, {"key": key, "value": result["value"], "warning": result["warning"]})
         except Exception as e:
             return _err(rid, 5001, str(e))
@@ -1368,25 +1511,12 @@ def _(rid, params: dict) -> dict:
                     nv = value
                 _save_cfg(cfg)
             elif key == "personality":
-                pname = value if value not in ("none", "default", "neutral") else ""
-                _write_config_key("display.personality", pname)
-                cfg = _load_cfg()
-                new_prompt = _resolve_personality_prompt(cfg)
-                _write_config_key("agent.system_prompt", new_prompt)
-                nv = value
                 sid_key = params.get("session_id", "")
-                if session:
-                    try:
-                        new_agent = _make_agent(sid_key, session["session_key"], session_id=session["session_key"])
-                        session["agent"] = new_agent
-                        with session["history_lock"]:
-                            session["history"] = []
-                            session["history_version"] = int(session.get("history_version", 0)) + 1
-                    except Exception:
-                        if session.get("agent"):
-                            agent = session["agent"]
-                            agent.ephemeral_system_prompt = new_prompt or None
-                            agent._cached_system_prompt = None
+                pname, new_prompt = _validate_personality(str(value or ""), cfg)
+                _write_config_key("display.personality", pname)
+                _write_config_key("agent.system_prompt", new_prompt)
+                nv = str(value or "default")
+                history_reset, info = _apply_personality_to_session(sid_key, session, new_prompt)
             else:
                 _write_config_key(f"display.{key}", value)
                 nv = value
@@ -1394,7 +1524,9 @@ def _(rid, params: dict) -> dict:
                     _emit("skin.changed", "", resolve_skin())
             resp = {"key": key, "value": nv}
             if key == "personality":
-                resp["cleared"] = True
+                resp["history_reset"] = history_reset
+                if info is not None:
+                    resp["info"] = info
             return _ok(rid, resp)
         except Exception as e:
             return _err(rid, 5001, str(e))
@@ -1425,6 +1557,11 @@ def _(rid, params: dict) -> dict:
         return _ok(rid, {"value": _load_cfg().get("display", {}).get("skin", "default")})
     if key == "personality":
         return _ok(rid, {"value": _load_cfg().get("display", {}).get("personality", "default")})
+    if key == "reasoning":
+        cfg = _load_cfg()
+        effort = str(cfg.get("agent", {}).get("reasoning_effort", "medium") or "medium")
+        display = "show" if bool(cfg.get("display", {}).get("show_reasoning", False)) else "hide"
+        return _ok(rid, {"value": effort, "display": display})
     if key == "mtime":
         cfg_path = _hermes_home / "config.yaml"
         try:
@@ -1510,14 +1647,15 @@ def _(rid, params: dict) -> dict:
             cat_map[cat].append([name, desc])
 
         skill_count = 0
+        warning = ""
         try:
             from agent.skill_commands import scan_skill_commands
             for k, info in sorted(scan_skill_commands().items()):
                 d = str(info.get("description", "Skill"))
                 all_pairs.append([k, d[:120] + ("…" if len(d) > 120 else "")])
                 skill_count += 1
-        except Exception:
-            pass
+        except Exception as e:
+            warning = f"skill discovery unavailable: {e}"
 
         for cat in cat_order:
             categories.append({"name": cat, "pairs": cat_map[cat]})
@@ -1529,6 +1667,7 @@ def _(rid, params: dict) -> dict:
             "canon": canon,
             "categories": categories,
             "skill_count": skill_count,
+            "warning": warning,
         })
     except Exception as e:
         return _err(rid, 5020, str(e))
@@ -1611,7 +1750,10 @@ def _(rid, params: dict) -> dict:
         qc = qcmds[name]
         if qc.get("type") == "exec":
             r = subprocess.run(qc.get("command", ""), shell=True, capture_output=True, text=True, timeout=30)
-            return _ok(rid, {"type": "exec", "output": (r.stdout or r.stderr)[:4000]})
+            output = ((r.stdout or "") + ("\n" if r.stdout and r.stderr else "") + (r.stderr or "")).strip()[:4000]
+            if r.returncode != 0:
+                return _err(rid, 4018, output or f"quick command failed with exit code {r.returncode}")
+            return _ok(rid, {"type": "exec", "output": output})
         if qc.get("type") == "alias":
             return _ok(rid, {"type": "alias", "target": qc.get("target", "")})
 
@@ -1692,15 +1834,18 @@ def _(rid, params: dict) -> dict:
             prefix_tag = ""
             path_part = query if not is_context else query
 
-        expanded = os.path.expanduser(path_part)
+        expanded = _normalize_completion_path(path_part)
         if expanded.endswith("/"):
             search_dir, match = expanded, ""
         else:
             search_dir = os.path.dirname(expanded) or "."
             match = os.path.basename(expanded)
 
+        if not os.path.isdir(search_dir):
+            return _ok(rid, {"items": []})
+
         match_lower = match.lower()
-        for entry in sorted(os.listdir(search_dir))[:200]:
+        for entry in sorted(os.listdir(search_dir)):
             if match and not entry.lower().startswith(match_lower):
                 continue
             if is_context and not prefix_tag and entry.startswith("."):
@@ -1725,8 +1870,8 @@ def _(rid, params: dict) -> dict:
             items.append({"text": text, "display": entry + suffix, "meta": "dir" if is_dir else ""})
             if len(items) >= 30:
                 break
-    except Exception:
-        pass
+    except Exception as e:
+        return _err(rid, 5021, str(e))
 
     return _ok(rid, {"items": items})
 
@@ -1742,39 +1887,83 @@ def _(rid, params: dict) -> dict:
         from prompt_toolkit.document import Document
         from prompt_toolkit.formatted_text import to_plain_text
 
-        completer = SlashCommandCompleter()
+        from agent.skill_commands import get_skill_commands
+
+        completer = SlashCommandCompleter(skill_commands_provider=lambda: get_skill_commands())
         doc = Document(text, len(text))
         items = [
             {"text": c.text, "display": c.display or c.text,
              "meta": to_plain_text(c.display_meta) if c.display_meta else ""}
             for c in completer.get_completions(doc, None)
         ][:30]
+        text_lower = text.lower()
+        extras = [
+            {"text": "/compact", "display": "/compact", "meta": "Toggle compact display mode"},
+            {"text": "/logs", "display": "/logs", "meta": "Show recent gateway log lines"},
+        ]
+        for extra in extras:
+            if extra["text"].startswith(text_lower) and not any(item["text"] == extra["text"] for item in items):
+                items.append(extra)
         return _ok(rid, {"items": items, "replace_from": text.rfind(" ") + 1 if " " in text else 1})
-    except Exception:
-        return _ok(rid, {"items": []})
+    except Exception as e:
+        return _err(rid, 5020, str(e))
+
+
+@method("model.options")
+def _(rid, params: dict) -> dict:
+    try:
+        from hermes_cli.model_switch import list_authenticated_providers
+        from hermes_cli.models import provider_model_ids
+
+        session = _sessions.get(params.get("session_id", ""))
+        agent = session.get("agent") if session else None
+        cfg = _load_cfg()
+        current_provider = getattr(agent, "provider", "") or ""
+        current_model = getattr(agent, "model", "") or _resolve_model()
+        providers = list_authenticated_providers(
+            current_provider=current_provider,
+            user_providers=cfg.get("providers") if isinstance(cfg.get("providers"), dict) else {},
+            custom_providers=cfg.get("custom_providers") if isinstance(cfg.get("custom_providers"), list) else [],
+            max_models=50,
+        )
+        for provider in providers:
+            try:
+                models = provider_model_ids(provider.get("slug"))
+                if models:
+                    provider["models"] = models
+                    provider["total_models"] = len(models)
+            except Exception as e:
+                provider["warning"] = f"model catalog unavailable: {e}"
+        return _ok(rid, {"providers": providers, "model": current_model, "provider": current_provider})
+    except Exception as e:
+        return _err(rid, 5033, str(e))
 
 
 # ── Methods: slash.exec ──────────────────────────────────────────────
 
 
-def _mirror_slash_side_effects(sid: str, session: dict, command: str):
+def _mirror_slash_side_effects(sid: str, session: dict, command: str) -> str:
     """Apply side effects that must also hit the gateway's live agent."""
     parts = command.lstrip("/").split(None, 1)
     if not parts:
-        return
+        return ""
     name, arg, agent = parts[0], (parts[1].strip() if len(parts) > 1 else ""), session.get("agent")
 
     try:
         if name == "model" and arg and agent:
-            _apply_model_switch(sid, session, arg)
-        elif name in ("personality", "prompt") and agent:
+            result = _apply_model_switch(sid, session, arg)
+            return result.get("warning", "")
+        elif name == "personality" and arg and agent:
+            _, new_prompt = _validate_personality(arg, _load_cfg())
+            _apply_personality_to_session(sid, session, new_prompt)
+        elif name == "prompt" and agent:
             cfg = _load_cfg()
             new_prompt = cfg.get("agent", {}).get("system_prompt", "") or ""
             agent.ephemeral_system_prompt = new_prompt or None
             agent._cached_system_prompt = None
         elif name == "compress" and agent:
             with session["history_lock"]:
-                _compress_session_history(session)
+                _compress_session_history(session, arg)
             _emit("session.info", sid, _session_info(agent))
         elif name == "fast" and agent:
             mode = arg.lower()
@@ -1788,8 +1977,9 @@ def _mirror_slash_side_effects(sid: str, session: dict, command: str):
         elif name == "stop":
             from tools.process_registry import ProcessRegistry
             ProcessRegistry().kill_all()
-    except Exception:
-        pass
+    except Exception as e:
+        return f"live session sync failed: {e}"
+    return ""
 
 
 @method("slash.exec")
@@ -1812,8 +2002,11 @@ def _(rid, params: dict) -> dict:
 
     try:
         output = worker.run(cmd)
-        _mirror_slash_side_effects(params.get("session_id", ""), session, cmd)
-        return _ok(rid, {"output": output or "(no output)"})
+        warning = _mirror_slash_side_effects(params.get("session_id", ""), session, cmd)
+        payload = {"output": output or "(no output)"}
+        if warning:
+            payload["warning"] = warning
+        return _ok(rid, payload)
     except Exception as e:
         try:
             worker.close()
@@ -1829,9 +2022,14 @@ def _(rid, params: dict) -> dict:
 def _(rid, params: dict) -> dict:
     action = params.get("action", "status")
     if action == "status":
-        return _ok(rid, {"enabled": os.environ.get("HERMES_VOICE", "0") == "1"})
+        env = os.environ.get("HERMES_VOICE", "").strip()
+        if env in {"0", "1"}:
+            return _ok(rid, {"enabled": env == "1"})
+        return _ok(rid, {"enabled": bool(_load_cfg().get("display", {}).get("voice_enabled", False))})
     if action in ("on", "off"):
-        os.environ["HERMES_VOICE"] = "1" if action == "on" else "0"
+        enabled = action == "on"
+        os.environ["HERMES_VOICE"] = "1" if enabled else "0"
+        _write_config_key("display.voice_enabled", enabled)
         return _ok(rid, {"enabled": action == "on"})
     return _err(rid, 4013, f"unknown voice action: {action}")
 
@@ -1965,12 +2163,34 @@ def _(rid, params: dict) -> dict:
         return _ok(rid, {"connected": bool(url), "url": url})
     if action == "connect":
         url = params.get("url", "http://localhost:9222")
-        os.environ["BROWSER_CDP_URL"] = url
         try:
+            import urllib.request
+            from urllib.parse import urlparse
             from tools.browser_tool import cleanup_all_browsers
+
+            parsed = urlparse(url if "://" in url else f"http://{url}")
+            if parsed.scheme not in {"http", "https", "ws", "wss"}:
+                return _err(rid, 4015, f"unsupported browser url: {url}")
+            probe_root = (
+                f"{'https' if parsed.scheme == 'wss' else 'http' if parsed.scheme == 'ws' else parsed.scheme}://{parsed.netloc}"
+            )
+            probe_urls = [f"{probe_root.rstrip('/')}/json/version", f"{probe_root.rstrip('/')}/json"]
+            ok = False
+            for probe in probe_urls:
+                try:
+                    with urllib.request.urlopen(probe, timeout=2.0) as resp:
+                        if 200 <= getattr(resp, "status", 200) < 300:
+                            ok = True
+                            break
+                except Exception:
+                    continue
+            if not ok:
+                return _err(rid, 5031, f"could not reach browser CDP at {url}")
+
+            os.environ["BROWSER_CDP_URL"] = url
             cleanup_all_browsers()
-        except Exception:
-            pass
+        except Exception as e:
+            return _err(rid, 5031, str(e))
         return _ok(rid, {"connected": True, "url": url})
     if action == "disconnect":
         os.environ.pop("BROWSER_CDP_URL", None)
@@ -1990,8 +2210,8 @@ def _(rid, params: dict) -> dict:
         return _ok(rid, {"plugins": [
             {"name": n, "version": getattr(i, "version", "?"), "enabled": getattr(i, "enabled", True)}
             for n, i in get_plugin_manager()._plugins.items()]})
-    except Exception:
-        return _ok(rid, {"plugins": []})
+    except Exception as e:
+        return _err(rid, 5032, str(e))
 
 
 @method("config.show")

ui-tui/.gitignore

@@ -1,3 +1,4 @@
 dist/
 node_modules/
 src/*.js
+docs/

ui-tui/src/components/modelPicker.tsx

@@ -0,0 +1,241 @@
+import { Box, Text, useInput } from '@hermes/ink'
+import { useEffect, useState } from 'react'
+
+import type { GatewayClient } from '../gatewayClient.js'
+import { asRpcResult, rpcErrorMessage } from '../lib/rpc.js'
+import type { Theme } from '../theme.js'
+
+interface ProviderItem {
+  is_current?: boolean
+  models?: string[]
+  name: string
+  slug: string
+  total_models?: number
+  warning?: string
+}
+
+const VISIBLE = 12
+
+const pageOffset = (count: number, sel: number) => Math.max(0, Math.min(sel - Math.floor(VISIBLE / 2), count - VISIBLE))
+
+export function ModelPicker({
+  gw,
+  onCancel,
+  onSelect,
+  sessionId,
+  t
+}: {
+  gw: GatewayClient
+  onCancel: () => void
+  onSelect: (value: string) => void
+  sessionId: string | null
+  t: Theme
+}) {
+  const [providers, setProviders] = useState<ProviderItem[]>([])
+  const [currentModel, setCurrentModel] = useState('')
+  const [err, setErr] = useState('')
+  const [loading, setLoading] = useState(true)
+  const [persistGlobal, setPersistGlobal] = useState(false)
+  const [providerIdx, setProviderIdx] = useState(0)
+  const [modelIdx, setModelIdx] = useState(0)
+  const [stage, setStage] = useState<'model' | 'provider'>('provider')
+
+  useEffect(() => {
+    gw.request('model.options', sessionId ? { session_id: sessionId } : {})
+      .then((raw: any) => {
+        const r = asRpcResult(raw)
+
+        if (!r) {
+          setErr('invalid response: model.options')
+          setLoading(false)
+
+          return
+        }
+
+        const next = (r.providers ?? []) as ProviderItem[]
+        setProviders(next)
+        setCurrentModel(String(r.model ?? ''))
+        setProviderIdx(
+          Math.max(
+            0,
+            next.findIndex(p => p.is_current)
+          )
+        )
+        setModelIdx(0)
+        setErr('')
+        setLoading(false)
+      })
+      .catch((e: unknown) => {
+        setErr(rpcErrorMessage(e))
+        setLoading(false)
+      })
+  }, [gw, sessionId])
+
+  const provider = providers[providerIdx]
+  const models = provider?.models ?? []
+
+  const visibleItems = (items: string[], sel: number) => {
+    const off = pageOffset(items.length, sel)
+
+    return { items: items.slice(off, off + VISIBLE), off }
+  }
+
+  useInput((ch, key) => {
+    if (key.escape) {
+      if (stage === 'model') {
+        setStage('provider')
+        setModelIdx(0)
+
+        return
+      }
+
+      onCancel()
+
+      return
+    }
+
+    const count = stage === 'provider' ? providers.length : models.length
+    const sel = stage === 'provider' ? providerIdx : modelIdx
+    const setSel = stage === 'provider' ? setProviderIdx : setModelIdx
+
+    if (key.upArrow && sel > 0) {
+      setSel(v => v - 1)
+
+      return
+    }
+
+    if (key.downArrow && sel < count - 1) {
+      setSel(v => v + 1)
+
+      return
+    }
+
+    if (key.return) {
+      if (stage === 'provider') {
+        if (!provider) {
+          return
+        }
+
+        setStage('model')
+        setModelIdx(0)
+
+        return
+      }
+
+      const model = models[modelIdx]
+
+      if (provider && model) {
+        onSelect(`${model} --provider ${provider.slug}${persistGlobal ? ' --global' : ''}`)
+      } else {
+        setStage('provider')
+      }
+
+      return
+    }
+
+    if (ch.toLowerCase() === 'g') {
+      setPersistGlobal(v => !v)
+
+      return
+    }
+
+    const n = ch === '0' ? 10 : parseInt(ch, 10)
+
+    if (!Number.isNaN(n) && n >= 1 && n <= Math.min(10, count)) {
+      const off = pageOffset(count, sel)
+
+      if (stage === 'provider') {
+        const next = off + n - 1
+
+        if (providers[next]) {
+          setProviderIdx(next)
+        }
+      } else if (provider && models[off + n - 1]) {
+        onSelect(`${models[off + n - 1]} --provider ${provider.slug}${persistGlobal ? ' --global' : ''}`)
+      }
+    }
+  })
+
+  if (loading) {
+    return <Text color={t.color.dim}>loading models…</Text>
+  }
+
+  if (err) {
+    return (
+      <Box flexDirection="column">
+        <Text color={t.color.label}>error: {err}</Text>
+        <Text color={t.color.dim}>Esc to cancel</Text>
+      </Box>
+    )
+  }
+
+  if (!providers.length) {
+    return (
+      <Box flexDirection="column">
+        <Text color={t.color.dim}>no authenticated providers</Text>
+        <Text color={t.color.dim}>Esc to cancel</Text>
+      </Box>
+    )
+  }
+
+  if (stage === 'provider') {
+    const rows = providers.map(
+      p => `${p.is_current ? '*' : ' '} ${p.name} · ${p.total_models ?? p.models?.length ?? 0} models`
+    )
+
+    const { items, off } = visibleItems(rows, providerIdx)
+
+    return (
+      <Box flexDirection="column">
+        <Text bold color={t.color.amber}>
+          Select Provider
+        </Text>
+        <Text color={t.color.dim}>Current model: {currentModel || '(unknown)'}</Text>
+        {provider?.warning ? <Text color={t.color.label}>warning: {provider.warning}</Text> : null}
+        {off > 0 && <Text color={t.color.dim}> ↑ {off} more</Text>}
+        {items.map((row, i) => {
+          const idx = off + i
+
+          return (
+            <Text color={providerIdx === idx ? t.color.cornsilk : t.color.dim} key={row}>
+              {providerIdx === idx ? '▸ ' : '  '}
+              {i + 1}. {row}
+            </Text>
+          )
+        })}
+        {off + VISIBLE < rows.length && <Text color={t.color.dim}> ↓ {rows.length - off - VISIBLE} more</Text>}
+        <Text color={t.color.dim}>persist: {persistGlobal ? 'global' : 'session'} · g toggle</Text>
+        <Text color={t.color.dim}>↑/↓ select · Enter choose · 1-9,0 quick · Esc cancel</Text>
+      </Box>
+    )
+  }
+
+  const { items, off } = visibleItems(models, modelIdx)
+
+  return (
+    <Box flexDirection="column">
+      <Text bold color={t.color.amber}>
+        Select Model
+      </Text>
+      <Text color={t.color.dim}>{provider?.name || '(unknown provider)'}</Text>
+      {!models.length ? <Text color={t.color.dim}>no models listed for this provider</Text> : null}
+      {provider?.warning ? <Text color={t.color.label}>warning: {provider.warning}</Text> : null}
+      {off > 0 && <Text color={t.color.dim}> ↑ {off} more</Text>}
+      {items.map((row, i) => {
+        const idx = off + i
+
+        return (
+          <Text color={modelIdx === idx ? t.color.cornsilk : t.color.dim} key={row}>
+            {modelIdx === idx ? '▸ ' : '  '}
+            {i + 1}. {row}
+          </Text>
+        )
+      })}
+      {off + VISIBLE < models.length && <Text color={t.color.dim}> ↓ {models.length - off - VISIBLE} more</Text>}
+      <Text color={t.color.dim}>persist: {persistGlobal ? 'global' : 'session'} · g toggle</Text>
+      <Text color={t.color.dim}>
+        {models.length ? '↑/↓ select · Enter switch · 1-9,0 quick · Esc back' : 'Enter/Esc back'}
+      </Text>
+    </Box>
+  )
+}

ui-tui/src/components/textInput.tsx

@@ -30,10 +30,68 @@ const dim = (s: string) => DIM + s + DIM_OFF
 let _seg: Intl.Segmenter | null = null
 const seg = () => (_seg ??= new Intl.Segmenter(undefined, { granularity: 'grapheme' }))
 
+function graphemeStops(s: string) {
+  const stops = [0]
+
+  for (const { index } of seg().segment(s)) {
+    if (index > 0) {
+      stops.push(index)
+    }
+  }
+
+  if (stops.at(-1) !== s.length) {
+    stops.push(s.length)
+  }
+
+  return stops
+}
+
+function snapPos(s: string, p: number) {
+  const pos = Math.max(0, Math.min(p, s.length))
+  let last = 0
+
+  for (const stop of graphemeStops(s)) {
+    if (stop > pos) {
+      break
+    }
+
+    last = stop
+  }
+
+  return last
+}
+
+function prevPos(s: string, p: number) {
+  const pos = snapPos(s, p)
+  let prev = 0
+
+  for (const stop of graphemeStops(s)) {
+    if (stop >= pos) {
+      return prev
+    }
+
+    prev = stop
+  }
+
+  return prev
+}
+
+function nextPos(s: string, p: number) {
+  const pos = snapPos(s, p)
+
+  for (const stop of graphemeStops(s)) {
+    if (stop > pos) {
+      return stop
+    }
+  }
+
+  return s.length
+}
+
 // ── Word movement ────────────────────────────────────────────────────
 
 function wordLeft(s: string, p: number) {
-  let i = p - 1
+  let i = snapPos(s, p) - 1
 
   while (i > 0 && /\s/.test(s[i]!)) {
     i--
@@ -47,7 +105,7 @@ function wordLeft(s: string, p: number) {
 }
 
 function wordRight(s: string, p: number) {
-  let i = p
+  let i = snapPos(s, p)
 
   while (i < s.length && !/\s/.test(s[i]!)) {
     i++
@@ -252,7 +310,7 @@ export function TextInput({
 
   const commit = (next: string, nextCur: number, track = true) => {
     const prev = vRef.current
-    const c = Math.max(0, Math.min(nextCur, next.length))
+    const c = snapPos(next, nextCur)
 
     if (track && next !== prev) {
       undo.current.push({ cursor: curRef.current, value: prev })
@@ -316,11 +374,10 @@ export function TextInput({
 
   useInput(
     (inp: string, k: Key, event: InputEvent) => {
-      // Some terminals normalize Ctrl+V to "v"; others deliver raw ^V (\x16).
-      const ctrlPaste = k.ctrl && (inp.toLowerCase() === 'v' || event.keypress.raw === '\x16')
-      const metaPaste = k.meta && inp.toLowerCase() === 'v'
+      const raw = event.keypress.raw
+      const metaPaste = raw === '\x1bv' || raw === '\x1bV'
 
-      if (ctrlPaste || metaPaste) {
+      if (metaPaste) {
         return void emitPaste({ cursor: curRef.current, hotkey: true, text: '', value: vRef.current })
       }
 
@@ -366,9 +423,9 @@ export function TextInput({
       } else if (k.end || (k.ctrl && inp === 'e')) {
         c = v.length
       } else if (k.leftArrow) {
-        c = mod ? wordLeft(v, c) : Math.max(0, c - 1)
+        c = mod ? wordLeft(v, c) : prevPos(v, c)
       } else if (k.rightArrow) {
-        c = mod ? wordRight(v, c) : Math.min(v.length, c + 1)
+        c = mod ? wordRight(v, c) : nextPos(v, c)
       } else if (k.meta && inp === 'b') {
         c = wordLeft(v, c)
       } else if (k.meta && inp === 'f') {
@@ -382,15 +439,16 @@ export function TextInput({
           v = v.slice(0, t) + v.slice(c)
           c = t
         } else {
-          v = v.slice(0, c - 1) + v.slice(c)
-          c--
+          const t = prevPos(v, c)
+          v = v.slice(0, t) + v.slice(c)
+          c = t
         }
       } else if (k.delete && fwdDel.current && c < v.length) {
         if (mod) {
           const t = wordRight(v, c)
           v = v.slice(0, c) + v.slice(t)
         } else {
-          v = v.slice(0, c) + v.slice(c + 1)
+          v = v.slice(0, c) + v.slice(nextPos(v, c))
         }
       } else if (k.ctrl && inp === 'w' && c > 0) {
         const t = wordLeft(v, c)

ui-tui/src/constants.ts

@@ -25,7 +25,7 @@ export const HOTKEYS: [string, string][] = [
   ['Ctrl+G', 'open $EDITOR for prompt'],
   ['Ctrl+L', 'new session (clear)'],
   ['Ctrl+T', 'cycle thinking detail'],
-  ['Ctrl+V / Alt+V', 'paste clipboard image'],
+  ['Alt+V / /paste', 'paste clipboard image'],
   ['Tab', 'apply completion'],
   ['↑/↓', 'completions / queue edit / history'],
   ['Ctrl+A/E', 'home / end of line'],

ui-tui/src/gatewayClient.ts

@@ -5,6 +5,8 @@ import { createInterface } from 'node:readline'
 
 const MAX_GATEWAY_LOG_LINES = 200
 const MAX_LOG_PREVIEW = 240
+const STARTUP_TIMEOUT_MS = Math.max(5000, parseInt(process.env.HERMES_TUI_STARTUP_TIMEOUT_MS ?? '15000', 10) || 15000)
+const REQUEST_TIMEOUT_MS = Math.max(30000, parseInt(process.env.HERMES_TUI_RPC_TIMEOUT_MS ?? '120000', 10) || 120000)
 
 export interface GatewayEvent {
   type: string
@@ -23,27 +25,78 @@ export class GatewayClient extends EventEmitter {
   private logs: string[] = []
   private pending = new Map<string, Pending>()
   private bufferedEvents: GatewayEvent[] = []
+  private pendingExit: number | null | undefined
+  private ready = false
+  private readyTimer: ReturnType<typeof setTimeout> | null = null
   private subscribed = false
+  private stdoutRl: ReturnType<typeof createInterface> | null = null
+  private stderrRl: ReturnType<typeof createInterface> | null = null
+
+  private publish(ev: GatewayEvent) {
+    if (ev.type === 'gateway.ready') {
+      this.ready = true
+
+      if (this.readyTimer) {
+        clearTimeout(this.readyTimer)
+        this.readyTimer = null
+      }
+    }
+
+    if (this.subscribed) {
+      this.emit('event', ev)
+
+      return
+    }
+
+    this.bufferedEvents.push(ev)
+  }
 
   start() {
     const root = process.env.HERMES_PYTHON_SRC_ROOT ?? resolve(import.meta.dirname, '../../')
+    const python = process.env.HERMES_PYTHON ?? resolve(root, 'venv/bin/python')
+    const cwd = process.env.HERMES_CWD || root
+    this.ready = false
+    this.pendingExit = undefined
+    this.stdoutRl?.close()
+    this.stderrRl?.close()
+    this.stdoutRl = null
+    this.stderrRl = null
 
-    this.proc = spawn(process.env.HERMES_PYTHON ?? resolve(root, 'venv/bin/python'), ['-m', 'tui_gateway.entry'], {
-      cwd: process.env.HERMES_CWD || root,
+    if (this.proc && !this.proc.killed && this.proc.exitCode === null) {
+      this.proc.kill()
+    }
+
+    if (this.readyTimer) {
+      clearTimeout(this.readyTimer)
+    }
+
+    this.readyTimer = setTimeout(() => {
+      if (this.ready) {
+        return
+      }
+
+      this.pushLog(`[startup] timed out waiting for gateway.ready (python=${python}, cwd=${cwd})`)
+      this.publish({ type: 'gateway.start_timeout', payload: { cwd, python } })
+    }, STARTUP_TIMEOUT_MS)
+
+    this.proc = spawn(python, ['-m', 'tui_gateway.entry'], {
+      cwd,
       stdio: ['pipe', 'pipe', 'pipe']
     })
 
-    createInterface({ input: this.proc.stdout! }).on('line', raw => {
+    this.stdoutRl = createInterface({ input: this.proc.stdout! })
+    this.stdoutRl.on('line', raw => {
       try {
         this.dispatch(JSON.parse(raw))
       } catch {
         const preview = raw.trim().slice(0, MAX_LOG_PREVIEW) || '(empty line)'
         this.pushLog(`[protocol] malformed stdout: ${preview}`)
-        this.emit('event', { type: 'gateway.protocol_error', payload: { preview } } satisfies GatewayEvent)
+        this.publish({ type: 'gateway.protocol_error', payload: { preview } } satisfies GatewayEvent)
       }
     })
 
-    createInterface({ input: this.proc.stderr! }).on('line', raw => {
+    this.stderrRl = createInterface({ input: this.proc.stderr! })
+    this.stderrRl.on('line', raw => {
       const line = raw.trim()
 
       if (!line) {
@@ -51,18 +104,28 @@ export class GatewayClient extends EventEmitter {
       }
 
       this.pushLog(line)
-      this.emit('event', { type: 'gateway.stderr', payload: { line } } satisfies GatewayEvent)
+      this.publish({ type: 'gateway.stderr', payload: { line } } satisfies GatewayEvent)
     })
 
     this.proc.on('error', err => {
       this.pushLog(`[spawn] ${err.message}`)
       this.rejectPending(new Error(`gateway error: ${err.message}`))
-      this.emit('event', { type: 'gateway.stderr', payload: { line: `[spawn] ${err.message}` } } satisfies GatewayEvent)
+      this.publish({ type: 'gateway.stderr', payload: { line: `[spawn] ${err.message}` } } satisfies GatewayEvent)
     })
 
     this.proc.on('exit', code => {
+      if (this.readyTimer) {
+        clearTimeout(this.readyTimer)
+        this.readyTimer = null
+      }
+
       this.rejectPending(new Error(`gateway exited${code === null ? '' : ` (${code})`}`))
-      this.emit('exit', code)
+
+      if (this.subscribed) {
+        this.emit('exit', code)
+      } else {
+        this.pendingExit = code
+      }
     })
   }
 
@@ -78,13 +141,7 @@ export class GatewayClient extends EventEmitter {
     }
 
     if (msg.method === 'event') {
-      const ev = msg.params as GatewayEvent
-
-      if (this.subscribed) {
-        this.emit('event', ev)
-      } else {
-        this.bufferedEvents.push(ev)
-      }
+      this.publish(msg.params as GatewayEvent)
     }
   }
 
@@ -110,6 +167,12 @@ export class GatewayClient extends EventEmitter {
     for (const ev of pending) {
       this.emit('event', ev)
     }
+
+    if (this.pendingExit !== undefined) {
+      const code = this.pendingExit
+      this.pendingExit = undefined
+      this.emit('exit', code)
+    }
   }
 
   getLogTail(limit = 20): string {
@@ -117,6 +180,10 @@ export class GatewayClient extends EventEmitter {
   }
 
   request(method: string, params: Record<string, unknown> = {}): Promise<unknown> {
+    if (!this.proc?.stdin || this.proc.killed || this.proc.exitCode !== null) {
+      this.start()
+    }
+
     if (!this.proc?.stdin) {
       return Promise.reject(new Error('gateway not running'))
     }
@@ -128,7 +195,7 @@ export class GatewayClient extends EventEmitter {
         if (this.pending.delete(id)) {
           reject(new Error(`timeout: ${method}`))
         }
-      }, 30_000)
+      }, REQUEST_TIMEOUT_MS)
 
       this.pending.set(id, {
         reject: e => {

ui-tui/src/hooks/useCompletion.ts

@@ -2,7 +2,7 @@ import { startTransition, useEffect, useRef, useState } from 'react'
 
 import type { GatewayClient } from '../gatewayClient.js'
 
-const TAB_PATH_RE = /((?:\.\.?\/|~\/|\/|@)[^\s]*)$/
+const TAB_PATH_RE = /((?:["']?(?:[A-Za-z]:[\\/]|\.{1,2}\/|~\/|\/|@|[^"'`\s]+\/))[^\s]*)$/
 
 export function useCompletion(input: string, blocked: boolean, gw: GatewayClient) {
   const [completions, setCompletions] = useState<{ text: string; display: string; meta: string }[]>([])
@@ -59,7 +59,18 @@ export function useCompletion(input: string, blocked: boolean, gw: GatewayClient
             setCompReplace(isSlash ? (r?.replace_from ?? 1) : input.length - (pathWord?.length ?? 0))
           })
         })
-        .catch(() => {})
+        .catch((e: unknown) => {
+          if (ref.current !== input) {
+            return
+          }
+
+          const meta = e instanceof Error && e.message ? e.message : 'unavailable'
+          startTransition(() => {
+            setCompletions([{ text: '', display: 'completion unavailable', meta }])
+            setCompIdx(0)
+            setCompReplace(isSlash ? 1 : input.length - (pathWord?.length ?? 0))
+          })
+        })
     }, 60)
 
     return () => clearTimeout(t)

ui-tui/src/lib/text.ts

@@ -243,4 +243,4 @@ export const userDisplay = (text: string): string => {
 }
 
 export const isPasteBackedText = (text: string): boolean =>
-  /\[\[paste:\d+(?:[^\n]*?)\]\]|\[paste #\d+ (?:attached|excerpt)\]/.test(text)
+  /\[\[paste:\d+(?:[^\n]*?)\]\]|\[paste #\d+ (?:attached|excerpt)(?:[^\n]*?)\]/.test(text)

ui-tui/src/types.ts

@@ -43,6 +43,7 @@ export interface SessionInfo {
   tools: Record<string, string[]>
   update_behind?: number | null
   update_command?: string
+  usage?: Usage
   version?: string
 }