refactor: extraer el router HTTP de LLM

tests/fixtures/architecture_wire_inventory.json

@@ -211,12 +211,6 @@
       "classification": "symbol_importable",
       "reason": "Extracts selected launch symbols with AST/exec; migrate to direct imports when that domain moves."
     },
-    {
-      "file": "tests/test_minimax_song_writer_prompt.py",
-      "target": "app/_launch_runtime.py",
-      "classification": "symbol_importable",
-      "reason": "Extracts selected launch symbols with AST/exec; migrate to direct imports when that domain moves."
-    },
     {
       "file": "tests/test_model_selection_persistence.py",
       "target": "app/_launch_runtime.py",

tests/fixtures/route_table.json

@@ -808,7 +808,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 80
     },
     {
@@ -818,7 +818,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 81
     },
     {
@@ -828,7 +828,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 82
     },
     {
@@ -838,7 +838,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 83
     },
     {
@@ -848,7 +848,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 84
     },
     {
@@ -858,7 +858,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 85
     },
     {
@@ -868,7 +868,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 86
     },
     {
@@ -878,7 +878,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 87
     },
     {
@@ -898,7 +898,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 89
     },
     {
@@ -908,7 +908,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 90
     },
     {
@@ -918,7 +918,7 @@
       "status_code": null,
       "response_model": null,
       "include_in_schema": null,
-      "source": "app/_launch_runtime.py",
+      "source": "app/routers/llm.py",
       "ordinal": 91
     },
     {

tests/test_h3_window_planner.py

@@ -380,14 +380,15 @@ def make_plan(*, golden: bool, dialogue: bool) -> dict:
     def test_ui_and_runtime_use_explicit_prompt_arrays(self):
         handler = (APP / "wgp.py").read_text(encoding="utf-8")
         launch = (APP / "_launch_runtime.py").read_text(encoding="utf-8")
+        llm_router = (APP / "routers" / "llm.py").read_text(encoding="utf-8")
         store = (ROOT / "ui" / "src" / "stores" / "useStore.ts").read_text(encoding="utf-8")
         advanced = (ROOT / "ui" / "src" / "components" / "Sidebar" / "AdvancedSettings.tsx").read_text(encoding="utf-8")
         prompt_input = (ROOT / "ui" / "src" / "components" / "Sidebar" / "PromptInput.tsx").read_text(encoding="utf-8")
         main_content = (ROOT / "ui" / "src" / "components" / "MainContent" / "MainContent.tsx").read_text(encoding="utf-8")
         guide = APP / "services" / "llm_guides" / "enhance" / "minimax_h3_sliding_windows.md"
         self.assertIn("h3_window_prompts=None", handler)
         self.assertIn("Using {len(prompts)} explicit", handler)
-        self.assertIn('/api/v1/llm/plan-h3-windows', launch)
+        self.assertIn('/api/v1/llm/plan-h3-windows', llm_router)
         self.assertIn("h3_window_plan_signature", launch)
         self.assertIn("api.planH3Windows", store)
         self.assertIn("Plan Prompt Across Windows", advanced)

tests/test_llm_router.py

@@ -0,0 +1,117 @@
+"""ASGI contracts for the extracted /api/v1/llm FastAPI routers."""
+
+from __future__ import annotations
+
+from unittest.mock import patch
+
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from routers.llm import create_llm_prompt_router, create_llm_router
+
+
+LLM_HTTP_SURFACE = [
+    ("GET", "/api/v1/llm/status", "llm_status"),
+    ("POST", "/api/v1/llm/load", "llm_load"),
+    ("POST", "/api/v1/llm/unload", "llm_unload"),
+    ("GET", "/api/v1/llm/models", "list_llm_models"),
+    ("GET", "/api/v1/llm/stream-status", "llm_stream_status"),
+    ("POST", "/api/v1/llm/generate", "llm_generate"),
+    ("POST", "/api/v1/llm/test", "llm_test"),
+    ("POST", "/api/v1/llm/write-song", "llm_write_song"),
+]
+
+LLM_PROMPT_HTTP_SURFACE = [
+    ("POST", "/api/v1/llm/plan-h3-windows", "llm_plan_h3_windows"),
+    ("POST", "/api/v1/llm/enhance-prompt", "llm_enhance_prompt"),
+    ("POST", "/api/v1/llm/describe-image", "llm_describe_image"),
+]
+
+
+def _core_router(**overrides):
+    kwargs = dict(
+        get_services_config=lambda: {"llm_device": "cpu"},
+        effective_llm_routing=lambda _services: ("local", "test-model", ""),
+        llm_provider_credentials=lambda _provider, _services, remote_url="": ("", remote_url),
+        llm_default_device=lambda: "cpu",
+        default_llm_repo="test-repo",
+        ensure_llm_loaded=lambda: None,
+        comic_writing_llm=lambda _body: None,
+    )
+    kwargs.update(overrides)
+    return create_llm_router(**kwargs)
+
+
+def _prompt_router(**overrides):
+    kwargs = dict(
+        get_services_config=lambda: {"nsfw_mode": False},
+        effective_llm_routing=lambda _services: ("local", "test-model", ""),
+        public_llm_providers=frozenset({"openai", "anthropic", "minimax", "grok", "deepseek"}),
+        ensure_llm_loaded=lambda: None,
+        get_model_def=lambda _model_type: {"architecture": "minimax_h3", "fps": 24},
+        get_lora_dir=lambda _model_type: "/tmp",
+        get_cached_hardware=lambda: {"gpu_vram_gb": 0.0},
+        get_enhancer_enabled=lambda: 0,
+        enhance_with_wangp=lambda *_args, **_kwargs: {"original": "x", "enhanced": "y"},
+    )
+    kwargs.update(overrides)
+    return create_llm_prompt_router(**kwargs)
+
+
+def _route_surface(router):
+    found = []
+    for route in router.routes:
+        methods = sorted(
+            method for method in (route.methods or set()) if method not in {"HEAD", "OPTIONS"}
+        )
+        for method in methods:
+            found.append((method, route.path, route.endpoint.__name__))
+    return found
+
+
+def test_llm_router_exposes_the_extracted_http_surface():
+    assert _route_surface(_core_router()) == LLM_HTTP_SURFACE
+
+
+def test_llm_prompt_router_exposes_the_extracted_http_surface():
+    assert _route_surface(_prompt_router()) == LLM_PROMPT_HTTP_SURFACE
+
+
+def test_generate_and_write_song_require_their_payloads():
+    app = FastAPI()
+    app.include_router(_core_router())
+    client = TestClient(app)
+    assert client.post("/api/v1/llm/generate", json={}).json()["detail"] == "prompt is required"
+    assert client.post("/api/v1/llm/write-song", json={}).json()["detail"] == "description is required"
+
+
+def test_generate_returns_llm_text_without_loading_wgp():
+    app = FastAPI()
+    app.include_router(_core_router())
+    client = TestClient(app)
+    with patch("services.llm_service.generate", return_value="hello") as generate:
+        response = client.post("/api/v1/llm/generate", json={"prompt": "Say hi"})
+    assert response.status_code == 200
+    assert response.json() == {"text": "hello"}
+    generate.assert_called_once()
+
+
+def test_plan_h3_windows_rejects_non_h3_models():
+    app = FastAPI()
+    app.include_router(_prompt_router(get_model_def=lambda _model_type: {"architecture": "ltx2"}))
+    client = TestClient(app)
+    response = client.post("/api/v1/llm/plan-h3-windows", json={
+        "prompt": "Clark turns toward the truck",
+        "model_type": "ltx2",
+    })
+    assert response.status_code == 400
+    assert "MiniMax H3" in response.json()["detail"]
+
+
+def test_describe_image_requires_a_path():
+    app = FastAPI()
+    app.include_router(_prompt_router())
+    client = TestClient(app)
+    response = client.post("/api/v1/llm/describe-image", json={})
+    assert response.status_code == 400
+    assert response.json()["detail"] == "image_path is required"

tests/test_minimax_h3.py

@@ -29,6 +29,7 @@
 _INT8_CONVROT_PATH = _APP / "shared" / "qtypes" / "int8_convrot.py"
 _WGP_PATH = _APP / "wgp.py"
 _LAUNCH_PATH = _APP / "_launch_runtime.py"
+_LLM_ROUTER_PATH = _APP / "routers" / "llm.py"
 _LLM_SERVICE_PATH = _APP / "services" / "llm_service.py"
 _DEFAULT_PATH = _APP / "defaults" / "minimax_h3.json"
 _LEGACY_DEFAULT_PATH = _APP / "defaults" / "minimax_h3_legacy.json"
@@ -945,10 +946,10 @@ def test_h3_prompt_guides_cover_native_audio_and_director(self):
         self.assertIn("proper names", ref2va_dialect_guide)
 
     def test_h3_enhance_path_preserves_context_ir_contract(self):
-        launch = _read(_LAUNCH_PATH)
+        llm_router = _read(_LLM_ROUTER_PATH)
         llm_service = _read(_LLM_SERVICE_PATH)
-        self.assertIn("needs_h3_context_ir", launch)
-        self.assertIn("enhancer_enabled > 0 and not needs_h3_context_ir", launch)
+        self.assertIn("needs_h3_context_ir", llm_router)
+        self.assertIn("enhancer_enabled > 0 and not needs_h3_context_ir", llm_router)
         self.assertIn("is_h3_context_ir", llm_service)
         self.assertIn("is_h3_ref2va", llm_service)
         self.assertIn("is_h3_structured = is_h3_context_ir or is_h3_ref2va", llm_service)

tests/test_minimax_song_writer_prompt.py

@@ -2,31 +2,15 @@
 
 from __future__ import annotations
 
-import ast
-import re
 import unittest
-from pathlib import Path
-
-
-def _load_functions(*names: str):
-    source = Path(__file__).parents[1].joinpath("app", "_launch_runtime.py").read_text(encoding="utf-8")
-    tree = ast.parse(source)
-    selected = [
-        node for node in tree.body
-        if isinstance(node, ast.FunctionDef) and node.name in names
-    ]
-    namespace = {"re": re}
-    exec(compile(ast.Module(body=selected, type_ignores=[]), "_launch_runtime.py", "exec"), namespace)
-    return tuple(namespace[name] for name in names)
-
-
-_minimax_song_request_prompt, _normalize_minimax_song_output, _parse_song_output, _parse_lyria_output, _optional_lyria_warning, _ace_song_request_prompt = _load_functions(
-    "_minimax_song_request_prompt",
-    "_normalize_minimax_song_output",
-    "_parse_song_output",
-    "_parse_lyria_output",
-    "_optional_lyria_warning",
-    "_ace_song_request_prompt",
+
+from routers.llm import (
+    _ace_song_request_prompt,
+    _minimax_song_request_prompt,
+    _normalize_minimax_song_output,
+    _optional_lyria_warning,
+    _parse_lyria_output,
+    _parse_song_output,
 )