fix(cli): honor env precedence for LLM and run config

Interactive selections and flag defaults overrode TRADINGAGENTS_* env vars.
Rule: an explicit env value or CLI flag wins; otherwise the env-applied
default is kept.

- Research depth: skip the prompt when both round-count env vars are set, and
  stop overwriting them (#977).
- Checkpoint: --checkpoint/--no-checkpoint is tri-state; omitting it keeps
  TRADINGAGENTS_CHECKPOINT_ENABLED (#976).
- Docker ollama: use TRADINGAGENTS_LLM_PROVIDER + OLLAMA_BASE_URL, not a bare
  LLM_PROVIDER the overlay never reads (#975).
- Reasoning/thinking knobs: settable via env; the prompt is skipped when set.
- Effort gating: forward effort only to models that accept it (Anthropic
  Opus 4.5+/Sonnet 4.6+, OpenAI reasoning models); drop it elsewhere.
- Boolean env values: raise a named error on invalid input instead of
  silently becoming False.

tradingagents/llm_clients/anthropic_client.py

@@ -12,20 +12,27 @@ _PASSTHROUGH_KWARGS = (
 )
 
 # Anthropic's extended-thinking ``effort`` parameter is accepted by Opus 4.5+
-# and Sonnet 4.5+ only. Haiku (any version shipped to date) 400s with
-# ``"This model does not support the effort parameter"`` (#831). Future
-# ``claude-{opus,sonnet}-X-Y`` releases inherit effort support via the
-# forward-compat pattern below; future Haiku stays excluded by default.
+# and Sonnet 4.6+ only. Sonnet 4.5 and any Haiku version 400 with
+# ``"This model does not support the effort parameter"`` (#831). The per-family
+# minimum version below is forward-compatible: future ``claude-{opus,sonnet}-X-Y``
+# releases inherit support automatically, while Sonnet 4.5 and Haiku stay excluded.
 _EFFORT_EXACT = {
     "claude-mythos-preview",  # non-standard preview name; effort-capable
 }
-_EFFORT_PATTERN = re.compile(r"^claude-(opus|sonnet)-\d+-\d+$")
+_EFFORT_MODEL = re.compile(r"^claude-(opus|sonnet)-(\d+)-(\d+)$")
+_EFFORT_MIN_VERSION = {"opus": (4, 5), "sonnet": (4, 6)}
 
 
 def _supports_effort(model: str) -> bool:
     """Whether Anthropic accepts the ``effort`` parameter for this model."""
     model_lc = model.lower()
-    return model_lc in _EFFORT_EXACT or bool(_EFFORT_PATTERN.match(model_lc))
+    if model_lc in _EFFORT_EXACT:
+        return True
+    match = _EFFORT_MODEL.match(model_lc)
+    if not match:
+        return False
+    family, major, minor = match.group(1), int(match.group(2)), int(match.group(3))
+    return (major, minor) >= _EFFORT_MIN_VERSION[family]
 
 
 class NormalizedChatAnthropic(ChatAnthropic):

tradingagents/llm_clients/openai_client.py

@@ -1,4 +1,5 @@
 import os
+import re
 from dataclasses import dataclass
 from typing import Any
 from urllib.parse import urlparse
@@ -150,6 +151,18 @@ _PASSTHROUGH_KWARGS = (
     "api_key", "callbacks", "http_client", "http_async_client",
 )
 
+# OpenAI's ``reasoning_effort`` is only accepted by reasoning models — the GPT-5
+# family and the o-series. Non-reasoning models (gpt-4.1, gpt-4o, ...) 400 with
+# "Unsupported parameter: 'reasoning.effort' is not supported with this model".
+# Drop the kwarg for those rather than crash the run.
+_OPENAI_REASONING_MODEL = re.compile(r"^(gpt-5|o[1-9])")
+
+
+def _supports_reasoning_effort(model: str) -> bool:
+    """Whether the (native OpenAI) model accepts ``reasoning_effort``."""
+    return bool(_OPENAI_REASONING_MODEL.match(model.lower().strip()))
+
+
 @dataclass(frozen=True)
 class ProviderSpec:
     """Declarative config for one OpenAI-compatible provider.
@@ -291,8 +304,11 @@ class OpenAIClient(BaseLLMClient):
 
         # Forward user-provided kwargs
         for key in _PASSTHROUGH_KWARGS:
-            if key in self.kwargs:
-                llm_kwargs[key] = self.kwargs[key]
+            if key not in self.kwargs:
+                continue
+            if key == "reasoning_effort" and not _supports_reasoning_effort(self.model):
+                continue
+            llm_kwargs[key] = self.kwargs[key]
 
         # The subclass (provider quirks) comes from the registry spec.
         return chat_cls(**llm_kwargs)