fix(tts): keep configured tts.provider authoritative; drop model schema override

tests/tools/test_tts_provider_authority.py

@@ -0,0 +1,74 @@
+"""Regression tests for model-boundary TTS provider authority.
+
+Programmatic callers may select a provider, but model-generated tool arguments
+must not override ``tts.provider``. Platform hints are advisory, so enforcement
+belongs in the registered schema and handler.
+"""
+
+import pytest
+
+from tools import tts_tool
+from tools.registry import registry
+
+
+class _ProviderProbe(Exception):
+    def __init__(self, provider: str):
+        super().__init__(provider)
+        self.provider = provider
+
+
+def test_programmatic_provider_override_remains_supported(monkeypatch):
+    """Trusted Python callers retain upstream's explicit provider-selection API."""
+    seen = []
+    monkeypatch.setattr(tts_tool, "_load_tts_config", lambda: {"provider": "xai"})
+    monkeypatch.setattr(
+        tts_tool,
+        "_resolve_max_text_length",
+        lambda provider, _config: seen.append(provider) or 4096,
+    )
+    monkeypatch.setattr(tts_tool, "_split_text_for_tts", lambda _text, _limit: [])
+
+    tts_tool.text_to_speech_tool("hello", provider="openai")
+
+    assert seen == ["openai"]
+
+
+def test_inner_dispatch_preserves_programmatic_provider_override(monkeypatch):
+    """The explicit provider must reach synthesis rather than reverting to config."""
+    monkeypatch.setattr(
+        tts_tool,
+        "_resolve_max_text_length",
+        lambda provider, _config: (_ for _ in ()).throw(_ProviderProbe(provider)),
+    )
+
+    with pytest.raises(_ProviderProbe) as caught:
+        tts_tool._text_to_speech_single(
+            "hello",
+            provider="openai",
+            tts_config_override={"provider": "xai"},
+        )
+
+    assert caught.value.provider == "openai"
+
+
+def test_registered_handler_discards_requested_provider(monkeypatch):
+    """Even a rogue payload must not cross the model-facing handler boundary."""
+    seen = {}
+    monkeypatch.setattr(
+        tts_tool,
+        "text_to_speech_tool",
+        lambda **kwargs: seen.update(kwargs) or "{}",
+    )
+
+    entry = registry.get_entry("text_to_speech")
+    assert entry is not None
+    entry.handler({"text": "hello", "provider": "openai"})
+
+    assert seen["provider"] is None
+
+
+def test_registered_model_schema_cannot_request_provider_override():
+    """Provider selection belongs to config, not model-generated tool arguments."""
+    entry = registry.get_entry("text_to_speech")
+    assert entry is not None
+    assert "provider" not in entry.schema["parameters"]["properties"]

tools/tts_tool.py

@@ -3521,7 +3521,7 @@ def text_to_speech_tool(
         output_path: Optional custom save path.
         speed: Optional playback speed multiplier (0.25-4.0).
         instructions: Optional voice-design guidance (tone, emotion, pacing).
-        provider: Optional TTS provider override.
+        provider: Optional TTS provider override for programmatic callers.
 
     Returns:
         str: JSON result with success, file_path, file_paths, and MEDIA tag.
@@ -3548,7 +3548,8 @@ def text_to_speech_tool(
         tts_config = dict(tts_config)  # shallow copy to avoid mutating the cache
         tts_config["speed"] = clamped
 
-    # Allow per-call provider override; fall back to the configured default.
+    # Allow per-call provider override for trusted programmatic callers; the
+    # model-facing registry handler deliberately never forwards this argument.
     if provider:
         provider = provider.lower().strip()
     else:
@@ -4518,18 +4519,8 @@ def _check(importer, label):
                 "description": (
                     "Optional voice-design guidance: tone, emotion, pacing, accent, "
                     "whispering, impressions (e.g. 'Speak in a cheerful, excited whisper'). "
-                    "Forwarded to the OpenAI backend (gpt-4o-mini-tts and OpenAI-compatible "
-                    "voice-design servers). Silently ignored by backends that don't support it."
-                )
-            },
-            "provider": {
-                "type": "string",
-                "description": (
-                    "Optional TTS provider override. Accepts built-in names "
-                    "(edge, openai, elevenlabs, minimax, xai, mistral, gemini, "
-                    "neutts, kittentts, piper), user-declared command provider "
-                    "names from tts.providers.<name>, or plugin-registered names. "
-                    "When omitted, the configured tts.provider from config.yaml is used."
+                    "Only used when the configured tts.provider supports voice-design "
+                    "instructions. Silently ignored by backends that don't support it."
                 )
             }
         },
@@ -4546,7 +4537,8 @@ def _check(importer, label):
         output_path=args.get("output_path"),
         speed=args.get("speed"),
         instructions=args.get("instructions"),
-        provider=args.get("provider")),
+        # provider deliberately not taken from model args — house config only
+        provider=None),
     check_fn=check_tts_requirements,
     emoji="🔊",
 )