diff --git a/evalharness/agent/envs/tau2_official.py b/evalharness/agent/envs/tau2_official.py
index 015352d..3151203 100644
--- a/evalharness/agent/envs/tau2_official.py
+++ b/evalharness/agent/envs/tau2_official.py
@@ -12,6 +12,7 @@ The env consumes a Sample whose metadata carries the official Task json
 import asyncio
 import json
 import os
+import sys
 from typing import Any, Dict, List, Optional
 
 from ...data.sample import ChatMessage, Sample
@@ -29,6 +30,18 @@ _TAU2_LOCAL_CANDIDATES = [
 ]
 
 
+def _silence_tau2_logs():
+    """tau2's loguru DEBUG dumps every full message to the terminal --
+    thousands of lines per run. Keep WARNING+ only."""
+    try:
+        from loguru import logger
+
+        logger.remove()
+        logger.add(sys.stderr, level='WARNING')
+    except Exception:
+        pass
+
+
 def _ensure_tau2_engine():
     """Auto-install the REAL tau2-bench engine when missing.
 
@@ -78,6 +91,7 @@ def _patch_tau2_generate(adapter, user_adapter=None, gen_kwargs=None) -> None:
         return
 
     _ensure_tau2_engine()
+    _silence_tau2_logs()
     import tau2.utils.llm_utils as llm_utils
 
     original = llm_utils.generate
@@ -136,7 +150,8 @@ def _patch_tau2_generate(adapter, user_adapter=None, gen_kwargs=None) -> None:
             calls.append(TauToolCall(id=c.id or f'call_{c.name}', name=c.name,
                                      arguments=args or {}))
         return AssistantMessage(role='assistant', content=text,
-                                tool_calls=calls or None, cost=None,
+                                tool_calls=calls or None,
+                                is_final_chunk=not calls, cost=None,
                                 usage=None, raw_data=None)
 
     # patch module attr AND every already-imported reference (official