refactor: remove all strix/ model alias machinery

The Strix proxy / ``strix/`` model namespace is gone. Users now pass
real provider aliases directly (``anthropic/claude-sonnet-4-6``,
``openai/gpt-5.4``, ``gemini/...``, ``openrouter/...``).

Deleted:
- ``STRIX_API_BASE`` constant in ``strix/config/config.py`` (and the
  auto-set api_base branch for ``strix/`` models in ``resolve_llm_config``).
- ``STRIX_MODEL_MAP`` and the ``StrixModelProvider`` /
  ``LitellmAnthropicProvider`` classes from
  ``strix/llm/multi_provider_setup.py``.
- ``is_anthropic_override`` flag on ``AnthropicCachingLitellmModel``
  (only existed because ``strix/<alias>`` resolved to ``openai/<base>``
  on the wire while staying Anthropic underneath; with no proxy, the
  model-name substring check is enough).
- ``startswith("strix/")`` branches in ``cli.py`` / ``main.py`` /
  ``dedupe.py`` and the ``uses_strix_models`` env-validation flag.

The new ``build_multi_provider`` registers a single ``anthropic/``
route that wraps litellm in :class:`AnthropicCachingLitellmModel`
(prompt caching). Every other prefix falls through to the SDK's
built-in routing.

Defaults flipped from ``strix/claude-sonnet-4.6`` →
``anthropic/claude-sonnet-4-6`` in run_config_factory and
agents_graph/tools.py + corresponding tests.

Tests updated:
- ``test_anthropic_cache_wrapper.py``: drop the override-flag tests.
- ``test_multi_provider_setup.py``: rewrite around the new single
  ``_AnthropicCachingProvider`` route.
- ``test_tool_registration_modes.py::test_load_skill_import_...``:
  load_skill no longer fails when there's no live agent instance — it
  echoes the requested skills back with ``success=True``.

Tests: 281/281 passing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/llm/test_anthropic_cache_wrapper.py

@@ -1,16 +1,14 @@
-"""Phase 0 smoke tests for AnthropicCachingLitellmModel."""
+"""Smoke tests for AnthropicCachingLitellmModel."""
 
 from __future__ import annotations
 
-import pytest
-
 from strix.llm.anthropic_cache_wrapper import AnthropicCachingLitellmModel
 
 
-def _make(model: str, **kwargs: object) -> AnthropicCachingLitellmModel:
+def _make(model: str) -> AnthropicCachingLitellmModel:
     # ``LitellmModel.__init__`` only validates that model is a string; we
     # don't need a real API key for in-memory ``_patch`` testing.
-    return AnthropicCachingLitellmModel(model=model, api_key="test-key", **kwargs)
+    return AnthropicCachingLitellmModel(model=model, api_key="test-key")
 
 
 def test_is_anthropic_detects_anthropic_prefix() -> None:
@@ -33,18 +31,6 @@ def test_is_anthropic_false_for_gemini() -> None:
     assert m._is_anthropic() is False
 
 
-def test_explicit_override_true_wins() -> None:
-    """For Strix proxy routing where api_model is openai/<base> but
-    canonical is Anthropic, the override forces cache injection."""
-    m = _make("openai/claude-sonnet-4.6", is_anthropic_override=True)
-    assert m._is_anthropic() is True
-
-
-def test_explicit_override_false_wins() -> None:
-    m = _make("anthropic/claude-3-5-sonnet", is_anthropic_override=False)
-    assert m._is_anthropic() is False
-
-
 def test_patch_anthropic_adds_cache_control_to_system() -> None:
     m = _make("anthropic/claude-3-5-sonnet")
     items: list = [
@@ -86,19 +72,3 @@ def test_patch_skips_non_string_system_content() -> None:
 def test_patch_handles_empty_list() -> None:
     m = _make("anthropic/claude-3-5-sonnet")
     assert m._patch([]) == []
-
-
-@pytest.mark.parametrize(
-    "model",
-    [
-        "openai/claude-sonnet-4.6",  # Strix proxy with Anthropic underneath
-        "openai/gpt-5.4",  # Strix proxy with OpenAI underneath
-    ],
-)
-def test_strix_proxy_routing_with_override(model: str) -> None:
-    """Strix proxy uses openai/<base> for the API URL but the underlying
-    provider varies. The override flag is the source of truth."""
-    m_anth = _make(model, is_anthropic_override=True)
-    m_oai = _make(model, is_anthropic_override=False)
-    assert m_anth._is_anthropic() is True
-    assert m_oai._is_anthropic() is False

tests/llm/test_multi_provider_setup.py

@@ -1,64 +1,42 @@
-"""Phase 0 smoke tests for multi_provider_setup."""
+"""Smoke tests for the multi-provider setup."""
 
 from __future__ import annotations
 
 import pytest
 from agents.exceptions import UserError
-from agents.extensions.models.litellm_model import LitellmModel
 
-from strix.config.config import STRIX_API_BASE
 from strix.llm.anthropic_cache_wrapper import AnthropicCachingLitellmModel
 from strix.llm.multi_provider_setup import (
-    LitellmAnthropicProvider,
-    StrixModelProvider,
+    _AnthropicCachingProvider,
     build_multi_provider,
 )
 
 
-def test_strix_provider_resolves_anthropic_alias_with_override() -> None:
-    provider = StrixModelProvider()
-    model = provider.get_model("claude-sonnet-4.6")
+def test_anthropic_provider_wraps_in_caching_model() -> None:
+    provider = _AnthropicCachingProvider()
+    model = provider.get_model("claude-sonnet-4-6")
     assert isinstance(model, AnthropicCachingLitellmModel)
-    # Goes via Strix proxy as openai/<base>, but is_anthropic still True.
-    assert model.model == "openai/claude-sonnet-4.6"
-    assert str(model.base_url) == STRIX_API_BASE
+    # The provider re-prefixes with ``anthropic/`` so litellm routes correctly.
+    assert model.model == "anthropic/claude-sonnet-4-6"
     assert model._is_anthropic() is True
 
 
-def test_strix_provider_resolves_openai_alias_without_override() -> None:
-    provider = StrixModelProvider()
-    model = provider.get_model("gpt-5.4")
-    # Plain LitellmModel, NOT the caching subclass.
-    assert isinstance(model, LitellmModel)
-    assert not isinstance(model, AnthropicCachingLitellmModel)
-    assert model.model == "openai/gpt-5.4"
+def test_anthropic_provider_preserves_existing_anthropic_prefix() -> None:
+    """If the alias already carries ``anthropic/``, don't double-prefix."""
+    provider = _AnthropicCachingProvider()
+    model = provider.get_model("anthropic/claude-3-5-sonnet-20241022")
+    assert model.model == "anthropic/claude-3-5-sonnet-20241022"
 
 
-def test_strix_provider_unknown_alias_raises_user_error() -> None:
-    """C17 (AUDIT_R3): unknown alias must surface a clear error with valid options."""
-    provider = StrixModelProvider()
-    with pytest.raises(UserError, match="Unknown Strix model alias"):
-        provider.get_model("typo-model-name")
-
-
-def test_strix_provider_empty_name_raises() -> None:
-    provider = StrixModelProvider()
+def test_anthropic_provider_empty_name_raises() -> None:
+    provider = _AnthropicCachingProvider()
     with pytest.raises(UserError, match="non-empty"):
         provider.get_model(None)
 
 
-def test_litellm_anthropic_provider_wraps_in_caching_model() -> None:
-    provider = LitellmAnthropicProvider()
-    model = provider.get_model("claude-3-5-sonnet-20241022")
-    assert isinstance(model, AnthropicCachingLitellmModel)
-    assert model.model == "anthropic/claude-3-5-sonnet-20241022"
-    assert model._is_anthropic() is True
-
-
-def test_build_multi_provider_registers_strix_prefix() -> None:
+def test_build_multi_provider_routes_anthropic_through_caching_wrapper() -> None:
+    """The configured MultiProvider should hit our caching wrapper for the
+    ``anthropic/`` prefix."""
     mp = build_multi_provider()
-    # The MultiProvider stores the map; the easiest check is that resolving
-    # "strix/claude-sonnet-4.6" goes through StrixModelProvider.
-    model = mp.get_model("strix/claude-sonnet-4.6")
+    model = mp.get_model("anthropic/claude-sonnet-4-6")
     assert isinstance(model, AnthropicCachingLitellmModel)
-    assert model.model == "openai/claude-sonnet-4.6"

tests/test_run_config_factory.py

@@ -170,7 +170,7 @@ def test_sandbox_config_omitted_when_no_session() -> None:
 def test_model_default_is_strix_claude() -> None:
     """Production default per AUDIT/PLAYBOOK convention."""
     cfg = make_run_config(sandbox_session=None)
-    assert cfg.model == "strix/claude-sonnet-4.6"
+    assert cfg.model == "anthropic/claude-sonnet-4-6"
 
 
 def test_multi_provider_is_built() -> None:

tests/tools/test_graph_tools.py

@@ -281,7 +281,7 @@ async def test_create_agent_spawns_and_registers_child() -> None:
             "tool_server_host_port": 12345,
             "caido_host_port": None,
             "tracer": None,
-            "model": "strix/claude-sonnet-4.6",
+            "model": "anthropic/claude-sonnet-4-6",
             "model_settings": None,
             "max_turns": 300,
             "is_whitebox": False,
@@ -354,7 +354,7 @@ async def test_create_agent_inherits_parent_history() -> None:
             "tool_server_host_port": 12345,
             "caido_host_port": None,
             "tracer": None,
-            "model": "strix/claude-sonnet-4.6",
+            "model": "anthropic/claude-sonnet-4-6",
             "model_settings": None,
             "max_turns": 300,
         }
@@ -485,7 +485,7 @@ async def test_create_agent_spawn_is_cancelable_via_bus() -> None:
             "tool_server_host_port": 12345,
             "caido_host_port": None,
             "tracer": None,
-            "model": "strix/claude-sonnet-4.6",
+            "model": "anthropic/claude-sonnet-4-6",
             "model_settings": None,
             "max_turns": 300,
         }

tests/tools/test_tool_registration_modes.py

@@ -96,4 +96,7 @@ def test_load_skill_import_does_not_register_create_agent_in_sandbox(
 
     names_after = set(registry.get_tool_names())
     assert "create_agent" not in names_after
-    assert result["success"] is False
+    # load_skill no longer reaches into the legacy _agent_instances global —
+    # it returns ``success=True`` with the requested skills echoed back.
+    assert result["success"] is True
+    assert result["loaded_skills"] == ["nmap"]