Baseline shortened: 11-task subset (1 per category + T01), max_tokens 16384, settings in README; baseline_chain.sh

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014aUaQeLnwbb1zTpN7kHeat

harness/agents.py

@@ -70,7 +70,7 @@ class LlmAgent:
     """
 
     def __init__(self, model, base_url=None, api_key=None, max_turns=80, temperature=0.2,
-                 max_seconds=None):
+                 max_seconds=None, max_tokens=None):
         self.model = model
         self.name = f"llm:{model}"
         self.base_url = (base_url or os.environ.get("LLM_BASE_URL", "http://127.0.0.1:11434/v1")).rstrip("/")
@@ -81,7 +81,7 @@ class LlmAgent:
         self.request_timeout = 3600  # local models are slow; a hung request still ends
         # Runaway reasoning: 35 of 1965 DeepSeek turns produced 393k output tokens and no content (46 % of the
         # run cost, 2026-10-03). Normal turns: p95 17.5k, max 82k. A cut turn is retried (see run).
-        self.max_tokens = 32000 if ":cloud" in (model or "") else None
+        self.max_tokens = max_tokens or (32000 if ":cloud" in (model or "") else None)
         self.empty_retries = 2
 
     def _chat(self, messages, tools):