roast: use chat API with system prompt + few-shot example for actual biting roasts

Switch from api/generate to api/chat so we can set a system role that
instructs the model to be genuinely savage. Add a few-shot example so
it knows what a roast looks like vs a backhanded compliment.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

matrixbot/commands.py

@@ -1777,22 +1777,33 @@ async def cmd_roast(client: AsyncClient, room_id: str, sender: str, args: str):
             lore = bio
             break
 
-    lore_clause = f" Here is some context about them: {lore}" if lore else ""
-    prompt = (
-        f"You are a comedy roast writer for a group of friends who all consented to being roasted for fun. "
-        f"Write a savage, witty, 1-2 sentence roast of a person named {display_name}.{lore_clause} "
-        f"Make it clever and specific. Output only the roast itself — no intro, no disclaimers, no apologies."
+    lore_clause = f"\nFacts about {display_name}: {lore}" if lore else ""
+    system_msg = (
+        "You are a savage comedy roast writer. Your job is to write brutal, funny, specific roasts. "
+        "A roast is NOT a compliment. It makes fun of the person's job, habits, appearance, or life choices. "
+        "Example of a good roast of a gamer: 'You've spent so many hours grinding XP you forgot to grind IRL — "
+        "congrats on hitting level 30 while your credit score is still level 1.' "
+        "Rules: output ONLY the roast, 1-2 sentences max, no softening, no disclaimers, no 'but seriously', "
+        "no compliments hidden in the roast. Be mean but funny."
     )
+    user_msg = f"Write a roast of {display_name}.{lore_clause}"
 
     try:
         timeout = aiohttp.ClientTimeout(total=30)
         async with aiohttp.ClientSession(timeout=timeout) as session:
             async with session.post(
-                f"{OLLAMA_URL}/api/generate",
-                json={"model": BALL_MODEL, "prompt": prompt, "stream": False},
+                f"{OLLAMA_URL}/api/chat",
+                json={
+                    "model": BALL_MODEL,
+                    "stream": False,
+                    "messages": [
+                        {"role": "system", "content": system_msg},
+                        {"role": "user", "content": user_msg},
+                    ],
+                },
             ) as response:
                 data = await response.json()
-                roast = data.get("response", "").strip()
+                roast = data.get("message", {}).get("content", "").strip()
                 if not roast:
                     raise ValueError("Empty roast response")
     except Exception as e: