Fix card truncation: lead JSON with effects+name, request concise flavor/art, bump card tokens

Root cause of pool-name + junk-effect cards: the model wrote name/flavor/long
art_prompt first and effects last, so max_tokens cut the JSON before it closed
-> parse failed -> whole card discarded (pool name, salvaged stray words like
'energy'). Now the requested JSON leads with effects then name (they survive a
cutoff), flavor/art_prompt are capped short, and card tokens raised to 320.

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

clients.py

@@ -51,7 +51,7 @@ def configure_mode() -> str:
         os.environ.setdefault("TABRAS_ART_MODEL", DEFAULT_ART_MODEL)
     os.environ.setdefault("TABRAS_AI_BOSS", "1")
     os.environ.setdefault("TABRAS_CARD_TEMPERATURE", "0.7")
-    os.environ.setdefault("TABRAS_CARD_MAX_TOKENS", "256")
+    os.environ.setdefault("TABRAS_CARD_MAX_TOKENS", "320")
     os.environ.setdefault("TABRAS_BOSS_MAX_TOKENS", "96")
     os.environ.setdefault("TABRAS_ART_STEPS", "4")
     return mode

generator.py

@@ -369,8 +369,11 @@ def llamacpp_card_prompt(payload: dict[str, Any], pack_cards: Sequence[dict[str,
     return "\n".join(
         (
             "Create exactly one Tabras card as JSON.",
-            "Shape: {\"name\": string, \"flavor\": string, \"art_prompt\": string, \"effects\": [{\"primitive_id\": string, \"weight\": 1}]}",
+            # effects + name come first so the card survives if the model is cut off
+            # mid-output; keep flavor and art_prompt short so the JSON always closes.
+            "Shape: {\"effects\": [{\"primitive_id\": string, \"weight\": 1}], \"name\": string, \"flavor\": string, \"art_prompt\": string}",
             "Use one or two effects. Do not write numeric damage, block, heal, or rules numbers.",
+            "Keep flavor under 12 words. Keep art_prompt under 18 words. Output only one compact JSON object.",
             "Fit the school identity, but invent the card's fantasy and tactical purpose freely.",
             "Do not describe Fire as a people or species; use fast damage plan or pressure plan.",
             "Prefer a new tactical purpose over repeating the current deck or pack.",
@@ -426,7 +429,8 @@ def llamacpp_retry_prompt(payload: dict[str, Any], need: str) -> str:
     return "\n".join(
         (
             "Create exactly one corrected Tabras card as JSON.",
-            "Shape: {\"name\": string, \"flavor\": string, \"art_prompt\": string, \"effects\": [{\"primitive_id\": string, \"weight\": 1}]}",
+            "Shape: {\"effects\": [{\"primitive_id\": string, \"weight\": 1}], \"name\": string, \"flavor\": string, \"art_prompt\": string}",
+            "Keep flavor under 12 words and art_prompt under 18 words; output one compact JSON object.",
             "Make flavor and art_prompt describe the same concrete scene.",
             "The art_prompt must describe an image scene only; no cards, text, frames, UI, or rules.",
             *naming_rules(),