fix(model): require confirmation for expensive model selections

Rebased onto current main and re-ported across the restructured
surfaces: model flows now thread confirm_provider/base_url/api_key
through hermes_cli/model_setup_flows.py, the Discord picker lives in
plugins/platforms/discord/adapter.py, and the web dashboard picker
applies chat-mode switches via config.set so the expensive-model
confirmation can ride the response.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

tests/hermes_cli/test_codex_models.py

@@ -133,7 +133,7 @@ def test_model_command_uses_runtime_access_token_for_codex_list(monkeypatch):
         captured["access_token"] = access_token
         return ["gpt-5.2-codex", "gpt-5.2"]
 
-    def _fake_prompt_model_selection(model_ids, current_model=""):
+    def _fake_prompt_model_selection(model_ids, current_model="", **_kwargs):
         captured["model_ids"] = list(model_ids)
         captured["current_model"] = current_model
         return None
@@ -181,7 +181,7 @@ def test_model_command_prompts_to_reuse_or_reauthenticate_codex_session(monkeypa
     )
     monkeypatch.setattr(
         "hermes_cli.auth._prompt_model_selection",
-        lambda model_ids, current_model="": None,
+        lambda model_ids, current_model="", **_kwargs: None,
     )
 
     _model_flow_openai_codex({}, current_model="gpt-5.4")
@@ -219,7 +219,7 @@ def test_model_command_uses_existing_codex_session_without_relogin(monkeypatch):
     )
     monkeypatch.setattr(
         "hermes_cli.auth._prompt_model_selection",
-        lambda model_ids, current_model="": None,
+        lambda model_ids, current_model="", **_kwargs: None,
     )
     monkeypatch.setattr(
         "hermes_cli.auth._login_openai_codex",

tests/hermes_cli/test_model_cost_guard.py

@@ -0,0 +1,97 @@
+from decimal import Decimal
+
+from agent.models_dev import ModelInfo
+from agent.usage_pricing import PricingEntry
+from hermes_cli.model_cost_guard import expensive_model_warning
+
+
+def test_no_warning_when_known_prices_are_at_threshold():
+    info = ModelInfo(
+        id="edge/model",
+        name="edge/model",
+        family="",
+        provider_id="test",
+        cost_input=20.0,
+        cost_output=100.0,
+    )
+
+    assert expensive_model_warning("edge/model", provider="test", model_info=info) is None
+
+
+def test_warns_when_models_dev_input_price_exceeds_threshold():
+    info = ModelInfo(
+        id="expensive/input",
+        name="expensive/input",
+        family="",
+        provider_id="test",
+        cost_input=20.01,
+        cost_output=1.0,
+    )
+
+    warning = expensive_model_warning(
+        "expensive/input",
+        provider="test",
+        model_info=info,
+    )
+
+    assert warning is not None
+    assert warning.input_cost_per_million == Decimal("20.01")
+    assert "EXPENSIVE MODEL WARNING" in warning.message
+    assert "$20/M input" in warning.message
+
+
+def test_warns_when_pricing_entry_output_price_exceeds_threshold(monkeypatch):
+    monkeypatch.setattr("agent.models_dev.get_model_info", lambda *_args, **_kwargs: None)
+    monkeypatch.setattr(
+        "agent.usage_pricing.get_pricing_entry",
+        lambda *_args, **_kwargs: PricingEntry(
+            input_cost_per_million=Decimal("1.00"),
+            output_cost_per_million=Decimal("100.01"),
+            source="provider_models_api",
+        ),
+    )
+
+    warning = expensive_model_warning("provider/expensive-output", provider="openrouter")
+
+    assert warning is not None
+    assert warning.output_cost_per_million == Decimal("100.01")
+    assert "$100.01/M" in warning.message
+
+
+def test_openai_gpt55_pro_adds_suggestion(monkeypatch):
+    monkeypatch.setattr("agent.models_dev.get_model_info", lambda *_args, **_kwargs: None)
+    monkeypatch.setattr(
+        "agent.usage_pricing.get_pricing_entry",
+        lambda *_args, **_kwargs: PricingEntry(
+            input_cost_per_million=Decimal("25"),
+            output_cost_per_million=Decimal("125"),
+            source="provider_models_api",
+        ),
+    )
+
+    warning = expensive_model_warning("openai/gpt-5.5-pro", provider="openrouter")
+
+    assert warning is not None
+    assert "did you mean to select openai/gpt-5.5?" in warning.message
+
+
+def test_openai_gpt55_pro_warns_for_nous_portal_pricing(monkeypatch):
+    monkeypatch.setattr("agent.models_dev.get_model_info", lambda *_args, **_kwargs: None)
+    monkeypatch.setattr(
+        "agent.usage_pricing.fetch_endpoint_model_metadata",
+        lambda base_url, api_key="": {
+            "openai/gpt-5.5-pro": {
+                "pricing": {
+                    "prompt": "0.000025",
+                    "completion": "0.000125",
+                }
+            }
+        },
+    )
+
+    warning = expensive_model_warning("openai/gpt-5.5-pro", provider="nous")
+
+    assert warning is not None
+    assert warning.input_cost_per_million == Decimal("25.000000")
+    assert warning.output_cost_per_million == Decimal("125.000000")
+    assert "did you mean to select openai/gpt-5.5?" in warning.message

tests/hermes_cli/test_model_picker_expensive_confirm.py

@@ -0,0 +1,64 @@
+from types import SimpleNamespace
+
+from hermes_cli.model_switch import ModelSwitchResult
+
+
+def _bound(fn, instance):
+    return fn.__get__(instance, type(instance))
+
+
+def test_prompt_toolkit_model_picker_defers_confirmation_off_key_handler(monkeypatch):
+    import cli as cli_mod
+
+    result = ModelSwitchResult(
+        success=True,
+        new_model="openai/gpt-5.5-pro",
+        target_provider="nous",
+    )
+    monkeypatch.setattr(
+        "hermes_cli.model_switch.switch_model",
+        lambda **_kwargs: result,
+    )
+
+    captured = {}
+
+    class _Thread:
+        def __init__(self, *, target, args, daemon):
+            captured["target"] = target
+            captured["args"] = args
+            captured["daemon"] = daemon
+
+        def start(self):
+            captured["started"] = True
+
+    monkeypatch.setattr(cli_mod.threading, "Thread", _Thread)
+
+    self_ = SimpleNamespace(
+        _app=object(),
+        _model_picker_state={
+            "stage": "model",
+            "provider_data": {"slug": "nous"},
+            "model_list": ["openai/gpt-5.5-pro"],
+            "selected": 0,
+            "user_provs": None,
+            "custom_provs": None,
+        },
+        provider="nous",
+        model="openai/gpt-5.5",
+        base_url="",
+        api_key="",
+        _restore_modal_input_snapshot=lambda: None,
+        _invalidate=lambda **_kwargs: None,
+    )
+    self_._close_model_picker = _bound(cli_mod.HermesCLI._close_model_picker, self_)
+    self_._confirm_and_apply_model_switch_result = (
+        lambda *_args: captured.setdefault("ran_inline", True)
+    )
+
+    _bound(cli_mod.HermesCLI._handle_model_picker_selection, self_)()
+
+    assert self_._model_picker_state is None
+    assert captured["started"] is True
+    assert captured["daemon"] is True
+    assert captured["args"] == (result, False)
+    assert "ran_inline" not in captured

tests/hermes_cli/test_terminal_menu_fallbacks.py

@@ -2,6 +2,7 @@
 cannot initialize (e.g. non-TTY, curses unavailable, terminal error)."""
 
 import subprocess
+from types import SimpleNamespace
 
 from hermes_cli.config import load_config, save_config
 
@@ -24,6 +25,46 @@ def test_prompt_model_selection_falls_back_on_menu_runtime_error(monkeypatch):
     assert selected == "model-b"
 
 
+def test_prompt_model_selection_requires_expensive_confirmation(monkeypatch, capsys):
+    from hermes_cli.auth import _prompt_model_selection
+
+    monkeypatch.setattr("hermes_cli.curses_ui.curses_radiolist", _raise_menu)
+    monkeypatch.setattr(
+        "hermes_cli.model_cost_guard.expensive_model_warning",
+        lambda *_args, **_kwargs: SimpleNamespace(message="EXPENSIVE MODEL WARNING"),
+    )
+    responses = iter(["1", "n"])
+    monkeypatch.setattr("builtins.input", lambda _prompt="": next(responses))
+
+    selected = _prompt_model_selection(
+        ["openai/gpt-5.5-pro"],
+        confirm_provider="nous",
+    )
+
+    out = capsys.readouterr().out
+    assert selected is None
+    assert "EXPENSIVE MODEL WARNING" in out
+
+
+def test_prompt_model_selection_allows_confirmed_expensive_model(monkeypatch):
+    from hermes_cli.auth import _prompt_model_selection
+
+    monkeypatch.setattr("hermes_cli.curses_ui.curses_radiolist", _raise_menu)
+    monkeypatch.setattr(
+        "hermes_cli.model_cost_guard.expensive_model_warning",
+        lambda *_args, **_kwargs: SimpleNamespace(message="EXPENSIVE MODEL WARNING"),
+    )
+    responses = iter(["1", "y"])
+    monkeypatch.setattr("builtins.input", lambda _prompt="": next(responses))
+
+    selected = _prompt_model_selection(
+        ["openai/gpt-5.5-pro"],
+        confirm_provider="nous",
+    )
+
+    assert selected == "openai/gpt-5.5-pro"
+
+
 def test_prompt_reasoning_effort_falls_back_on_menu_runtime_error(monkeypatch):
     from hermes_cli.main import _prompt_reasoning_effort_selection
 

tests/hermes_cli/test_web_server.py

@@ -4,6 +4,7 @@ import os
 import json
 import shutil
 from pathlib import Path
+from types import SimpleNamespace
 from unittest.mock import patch, MagicMock
 
 import pytest
@@ -1069,6 +1070,41 @@ class TestWebServerEndpoints:
         assert "GATEWAY_PROXY_URL" not in managed
         assert "GATEWAY_PROXY_URL" in _MESSAGING_KEYS_PAGE_KEYS
 
+    def test_model_set_requires_confirmation_for_expensive_model(self, monkeypatch):
+        monkeypatch.setattr(
+            "hermes_cli.model_cost_guard.expensive_model_warning",
+            lambda *_args, **_kwargs: SimpleNamespace(message="EXPENSIVE MODEL WARNING"),
+        )
+
+        resp = self.client.post(
+            "/api/model/set",
+            json={
+                "scope": "main",
+                "provider": "nous",
+                "model": "openai/gpt-5.5-pro",
+            },
+        )
+
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["ok"] is False
+        assert data["confirm_required"] is True
+        assert data["confirm_message"] == "EXPENSIVE MODEL WARNING"
+
+        confirmed = self.client.post(
+            "/api/model/set",
+            json={
+                "scope": "main",
+                "provider": "nous",
+                "model": "openai/gpt-5.5-pro",
+                "confirm_expensive_model": True,
+            },
+        )
+
+        assert confirmed.status_code == 200
+        assert confirmed.json()["ok"] is True
+
+
     def test_reveal_env_var(self, tmp_path):
         """POST /api/env/reveal should return the real unredacted value."""
         from hermes_cli.config import save_env_value