feat(2.0): W1 — fast-Spur ohne Thinking (flotte Antworten) + tool-call-Cap

Gateway injiziert chat_template_kwargs.enable_thinking=false fuer die fast-
Spur (Qwen3.6 ist Reasoning-Modell → sonst lahm/leer). heavy behaelt
Thinking. Env MC_FAST_NO_THINK. Hermes-Thrash war poisoned Persistent-
Session (fresh=clean, 34k statt 249k verifiziert); code_execution.max_tool_
calls 50->20 auf der Box (Historie unangetastet).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/config.py

@@ -52,7 +52,7 @@ PORT = int(os.environ.get("MC_PORT", "9000"))
 FRONTEND_DIST = Path(os.environ.get("MC_FRONTEND_DIST", str(Path(__file__).resolve().parent.parent / "frontend" / "dist")))
 
 # Version (Phase 0 — Greenfield-Skeleton).
-VERSION = "2.0.0-phase6"
+VERSION = "2.0.0-w1"
 
 # Gemeinsame YAML-Instanz (preserve_quotes hält Kommentare/Quotes in config.yaml).
 yaml = YAML()

backend/routers/gateway_proxy.py

@@ -12,7 +12,7 @@ from fastapi import APIRouter, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 
 from config import LLAMA_SWAP_URL
-from services.router_logic import choose_model
+from services.router_logic import FAST, FAST_NO_THINK, choose_model
 
 router = APIRouter(prefix="/v1")
 
@@ -32,7 +32,11 @@ async def _proxy(path: str, request: Request):
         body["model"] = alias
         routed = {"x-mc-routed-to": alias, "x-mc-route-reason": reason}
     else:
+        alias = requested
         routed = {"x-mc-routed-to": requested}
+    # fast-Spur: Thinking aus für flotte Antworten (sofern Client es nicht selbst setzt).
+    if FAST_NO_THINK and alias == FAST and "chat_template_kwargs" not in body:
+        body["chat_template_kwargs"] = {"enable_thinking": False}
     url = f"{LLAMA_SWAP_URL}{path}"
 
     if body.get("stream"):

backend/services/router_logic.py

@@ -9,6 +9,9 @@ import re
 FAST = os.environ.get("MC_ROUTE_FAST", "fast")
 HEAVY = os.environ.get("MC_ROUTE_HEAVY", "heavy")
 HEAVY_CHARS = int(os.environ.get("MC_GATEWAY_HEAVY_CHARS", "8000"))
+# Thinking auf der fast-Spur ausschalten → flotte Alltags-Antworten (Qwen3.6 ist ein
+# Reasoning-Modell). heavy behält Thinking für harte Aufgaben. Abschaltbar via Env.
+FAST_NO_THINK = os.environ.get("MC_FAST_NO_THINK", "1") not in ("0", "false", "")
 
 _HEAVY_KW = re.compile(
     r"\b(beweis|prove|theorem|refactor|architect|komplex|complex|schwierig|"