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.

.env.example

@@ -55,3 +55,8 @@ NVIDIA_API_KEY=
 # honor it). Unset leaves each provider at its default. See the README
 # "Reproducibility" note — no setting makes LLM output fully deterministic.
 #TRADINGAGENTS_TEMPERATURE=0.0
+# Provider-specific reasoning/thinking depth (optional; unset = provider
+# default). Setting one also skips the matching interactive prompt.
+#TRADINGAGENTS_OPENAI_REASONING_EFFORT=medium
+#TRADINGAGENTS_GOOGLE_THINKING_LEVEL=high
+#TRADINGAGENTS_ANTHROPIC_EFFORT=high

cli/main.py

@@ -517,6 +517,20 @@ def get_user_selections():
             box_content += f"\n[dim]Default: {default}[/dim]"
         return Panel(box_content, border_style="blue", padding=(1, 2))
 
+    def thinking_value_or_prompt(env_var, config_key, label, box_title, box_body, prompt_fn):
+        """Return the env-configured reasoning/thinking value, or prompt for it.
+
+        When ``env_var`` is set the interactive choice is skipped and the value
+        the env overlay placed on DEFAULT_CONFIG is used — mirroring the
+        env-precedence rule applied to the other selection steps.
+        """
+        if os.environ.get(env_var):
+            value = DEFAULT_CONFIG[config_key]
+            console.print(f"[green]✓ {label} from environment:[/green] {value}")
+            return value
+        console.print(create_question_box(box_title, box_body))
+        return prompt_fn()
+
     # Step 1: Ticker symbol
     console.print(
         create_question_box(
@@ -571,13 +585,27 @@ def get_user_selections():
         f"[green]Selected analysts:[/green] {', '.join(analyst.value for analyst in selected_analysts)}"
     )
 
-    # Step 5: Research depth
-    console.print(
-        create_question_box(
-            "Step 5: Research Depth", "Select your research depth level"
-        )
+    # Step 5: Research depth (skipped when both round counts are set via env).
+    # Research depth maps to the debate + risk round counts; when both are
+    # supplied through TRADINGAGENTS_MAX_DEBATE_ROUNDS / _MAX_RISK_ROUNDS we keep
+    # the run non-interactive and honor the env values (#977).
+    depth_from_env = bool(os.environ.get("TRADINGAGENTS_MAX_DEBATE_ROUNDS")) and bool(
+        os.environ.get("TRADINGAGENTS_MAX_RISK_ROUNDS")
     )
-    selected_research_depth = select_research_depth()
+    if depth_from_env:
+        selected_research_depth = DEFAULT_CONFIG["max_debate_rounds"]
+        console.print(
+            f"[green]✓ Research depth from environment:[/green] "
+            f"{DEFAULT_CONFIG['max_debate_rounds']} debate / "
+            f"{DEFAULT_CONFIG['max_risk_discuss_rounds']} risk rounds"
+        )
+    else:
+        console.print(
+            create_question_box(
+                "Step 5: Research Depth", "Select your research depth level"
+            )
+        )
+        selected_research_depth = select_research_depth()
 
     # Step 6: LLM Provider (skipped when set via TRADINGAGENTS_LLM_PROVIDER).
     # The backend URL comes from TRADINGAGENTS_LLM_BACKEND_URL when set,
@@ -649,43 +677,38 @@ def get_user_selections():
         selected_shallow_thinker = select_shallow_thinking_agent(selected_llm_provider)
         selected_deep_thinker = select_deep_thinking_agent(selected_llm_provider)
 
-    # Step 8: Provider-specific thinking configuration
+    # Step 8: Provider-specific reasoning/thinking configuration. Each knob is
+    # settable via its TRADINGAGENTS_* env var; when that var is set (or the
+    # provider itself came from env) the prompt is skipped and the configured
+    # value is used — same env-precedence rule as the steps above. None = each
+    # provider's own default.
     thinking_level = None
     reasoning_effort = None
     anthropic_effort = None
 
     provider_lower = selected_llm_provider.lower()
-    # When the provider is configured via environment we keep the run fully
-    # non-interactive and use the config defaults (None = each provider's own
-    # default reasoning/thinking behavior) instead of prompting.
     if provider_from_env:
         thinking_level = DEFAULT_CONFIG["google_thinking_level"]
         reasoning_effort = DEFAULT_CONFIG["openai_reasoning_effort"]
         anthropic_effort = DEFAULT_CONFIG["anthropic_effort"]
     elif provider_lower == "google":
-        console.print(
-            create_question_box(
-                "Step 8: Thinking Mode",
-                "Configure Gemini thinking mode"
-            )
+        thinking_level = thinking_value_or_prompt(
+            "TRADINGAGENTS_GOOGLE_THINKING_LEVEL", "google_thinking_level",
+            "Gemini thinking mode", "Step 8: Thinking Mode",
+            "Configure Gemini thinking mode", ask_gemini_thinking_config,
         )
-        thinking_level = ask_gemini_thinking_config()
     elif provider_lower == "openai":
-        console.print(
-            create_question_box(
-                "Step 8: Reasoning Effort",
-                "Configure OpenAI reasoning effort level"
-            )
+        reasoning_effort = thinking_value_or_prompt(
+            "TRADINGAGENTS_OPENAI_REASONING_EFFORT", "openai_reasoning_effort",
+            "Reasoning effort", "Step 8: Reasoning Effort",
+            "Configure OpenAI reasoning effort level", ask_openai_reasoning_effort,
         )
-        reasoning_effort = ask_openai_reasoning_effort()
     elif provider_lower == "anthropic":
-        console.print(
-            create_question_box(
-                "Step 8: Effort Level",
-                "Configure Claude effort level"
-            )
+        anthropic_effort = thinking_value_or_prompt(
+            "TRADINGAGENTS_ANTHROPIC_EFFORT", "anthropic_effort",
+            "Claude effort", "Step 8: Effort Level",
+            "Configure Claude effort level", ask_anthropic_effort,
         )
-        anthropic_effort = ask_anthropic_effort()
 
     return {
         "ticker": selected_ticker,
@@ -1019,14 +1042,20 @@ def format_tool_args(args, max_length=80) -> str:
         return result[:max_length - 3] + "..."
     return result
 
-def run_analysis(checkpoint: bool = False):
-    # First get all user selections
-    selections = get_user_selections()
+def _build_run_config(selections: dict, checkpoint: bool | None) -> dict:
+    """Assemble the run config from interactive selections, honoring env precedence.
 
-    # Create config with selected research depth
+    Round counts and checkpoint follow "explicit env/flag wins": an env-applied
+    value on DEFAULT_CONFIG is preserved unless the user overrode it on the CLI.
+    """
     config = DEFAULT_CONFIG.copy()
-    config["max_debate_rounds"] = selections["research_depth"]
-    config["max_risk_discuss_rounds"] = selections["research_depth"]
+    # Research depth sets both round counts, but an explicit env override
+    # (TRADINGAGENTS_MAX_DEBATE_ROUNDS / _MAX_RISK_ROUNDS) wins over the
+    # interactive selection — leave the env-applied value in place (#977).
+    if not os.environ.get("TRADINGAGENTS_MAX_DEBATE_ROUNDS"):
+        config["max_debate_rounds"] = selections["research_depth"]
+    if not os.environ.get("TRADINGAGENTS_MAX_RISK_ROUNDS"):
+        config["max_risk_discuss_rounds"] = selections["research_depth"]
     config["quick_think_llm"] = selections["shallow_thinker"]
     config["deep_think_llm"] = selections["deep_thinker"]
     config["backend_url"] = selections["backend_url"]
@@ -1036,7 +1065,18 @@ def run_analysis(checkpoint: bool = False):
     config["openai_reasoning_effort"] = selections.get("openai_reasoning_effort")
     config["anthropic_effort"] = selections.get("anthropic_effort")
     config["output_language"] = selections.get("output_language", "English")
-    config["checkpoint_enabled"] = checkpoint
+    # --checkpoint/--no-checkpoint overrides only when explicitly given; omitting
+    # the flag preserves TRADINGAGENTS_CHECKPOINT_ENABLED / the default (#976).
+    if checkpoint is not None:
+        config["checkpoint_enabled"] = checkpoint
+    return config
+
+
+def run_analysis(checkpoint: bool | None = None):
+    # First get all user selections
+    selections = get_user_selections()
+
+    config = _build_run_config(selections, checkpoint)
 
     # Create stats callback handler for tracking LLM/tool calls
     stats_handler = StatsCallbackHandler()
@@ -1316,10 +1356,11 @@ def run_analysis(checkpoint: bool = False):
 
 @app.command()
 def analyze(
-    checkpoint: bool = typer.Option(
-        False,
-        "--checkpoint",
-        help="Enable checkpoint/resume: save state after each node so a crashed run can resume.",
+    checkpoint: bool | None = typer.Option(
+        None,
+        "--checkpoint/--no-checkpoint",
+        help="Enable/disable checkpoint-resume (save state after each node so a "
+        "crashed run can resume). Omit to honor TRADINGAGENTS_CHECKPOINT_ENABLED.",
     ),
     clear_checkpoints: bool = typer.Option(
         False,

docker-compose.yml

@@ -20,7 +20,8 @@ services:
     env_file:
       - .env
     environment:
-      - LLM_PROVIDER=ollama
+      - TRADINGAGENTS_LLM_PROVIDER=ollama
+      - OLLAMA_BASE_URL=http://ollama:11434/v1
     volumes:
       - tradingagents_data:/home/appuser/.tradingagents
     depends_on:

tests/test_anthropic_effort.py

@@ -1,9 +1,9 @@
 """Tests for Anthropic effort-parameter gating (#831).
 
-Haiku 4.5 (and current Haiku versions) reject the ``effort`` parameter
-with a 400. Opus 4.5+ and Sonnet 4.5+ accept it. The gate uses a
-forward-compat regex so future ``claude-{opus,sonnet}-X-Y`` releases
-inherit support automatically.
+Haiku (any version) and Sonnet 4.5 reject the ``effort`` parameter with a
+400. Only Opus 4.5+ and Sonnet 4.6+ accept it. The gate uses a per-family
+minimum version so future ``claude-{opus,sonnet}-X-Y`` releases inherit
+support automatically.
 """
 
 import pytest
@@ -24,9 +24,13 @@ def _capture_kwargs(monkeypatch):
 class TestEffortGate:
     @pytest.mark.parametrize(
         "model",
-        ["claude-haiku-4-5", "claude-haiku-5-0", "claude-haiku-4-7-preview"],
+        [
+            "claude-haiku-4-5", "claude-haiku-5-0", "claude-haiku-4-7-preview",
+            # Sonnet 4.5 (and earlier) 400 on effort — only Sonnet 4.6+ supports it.
+            "claude-sonnet-4-5", "claude-sonnet-4-0",
+        ],
     )
-    def test_haiku_does_not_receive_effort(self, monkeypatch, model):
+    def test_unsupported_models_do_not_receive_effort(self, monkeypatch, model):
         captured = _capture_kwargs(monkeypatch)
         mod.AnthropicClient(model=model, effort="medium", api_key="x").get_llm()
         assert "effort" not in captured["kwargs"]
@@ -35,7 +39,7 @@ class TestEffortGate:
         "model",
         [
             "claude-opus-4-5", "claude-opus-4-6", "claude-opus-4-7",
-            "claude-sonnet-4-5", "claude-sonnet-4-6",
+            "claude-sonnet-4-6",
         ],
     )
     def test_current_opus_and_sonnet_receive_effort(self, monkeypatch, model):

tests/test_cli_config_precedence.py

@@ -0,0 +1,69 @@
+"""CLI config precedence (#976, #977).
+
+An explicit environment override for the debate/risk round counts, or the
+checkpoint flag, must win over the interactive research-depth selection — the CLI
+must not clobber an env-configured value back to a prompt/flag default.
+"""
+
+from unittest import mock
+
+import pytest
+
+import cli.main as m
+
+# Minimal selections dict shaped like get_user_selections()'s return value.
+SELECTIONS = {
+    "research_depth": 5,
+    "shallow_thinker": "gpt-5.4-mini",
+    "deep_thinker": "gpt-5.5",
+    "backend_url": None,
+    "llm_provider": "openai",
+    "google_thinking_level": None,
+    "openai_reasoning_effort": None,
+    "anthropic_effort": None,
+    "output_language": "English",
+}
+
+
+def test_research_depth_sets_both_rounds_without_env(monkeypatch):
+    for var in ("TRADINGAGENTS_MAX_DEBATE_ROUNDS", "TRADINGAGENTS_MAX_RISK_ROUNDS"):
+        monkeypatch.delenv(var, raising=False)
+    cfg = m._build_run_config(SELECTIONS, checkpoint=None)
+    assert cfg["max_debate_rounds"] == 5
+    assert cfg["max_risk_discuss_rounds"] == 5
+
+
+def test_env_round_counts_win_over_selection(monkeypatch):
+    monkeypatch.setenv("TRADINGAGENTS_MAX_DEBATE_ROUNDS", "2")
+    monkeypatch.setenv("TRADINGAGENTS_MAX_RISK_ROUNDS", "4")
+    # DEFAULT_CONFIG already reflects the env (applied at import); emulate that.
+    patched = dict(m.DEFAULT_CONFIG, max_debate_rounds=2, max_risk_discuss_rounds=4)
+    with mock.patch.object(m, "DEFAULT_CONFIG", patched):
+        cfg = m._build_run_config(SELECTIONS, checkpoint=None)
+    assert cfg["max_debate_rounds"] == 2  # env value, not research_depth=5
+    assert cfg["max_risk_discuss_rounds"] == 4
+
+
+def test_partial_env_only_overrides_that_count(monkeypatch):
+    monkeypatch.setenv("TRADINGAGENTS_MAX_DEBATE_ROUNDS", "2")
+    monkeypatch.delenv("TRADINGAGENTS_MAX_RISK_ROUNDS", raising=False)
+    patched = dict(m.DEFAULT_CONFIG, max_debate_rounds=2)
+    with mock.patch.object(m, "DEFAULT_CONFIG", patched):
+        cfg = m._build_run_config(SELECTIONS, checkpoint=None)
+    assert cfg["max_debate_rounds"] == 2  # env wins
+    assert cfg["max_risk_discuss_rounds"] == 5  # falls through to research_depth
+
+
+def test_checkpoint_none_preserves_env_default():
+    patched = dict(m.DEFAULT_CONFIG, checkpoint_enabled=True)  # e.g. env-enabled
+    with mock.patch.object(m, "DEFAULT_CONFIG", patched):
+        cfg = m._build_run_config(SELECTIONS, checkpoint=None)
+    assert cfg["checkpoint_enabled"] is True  # not clobbered back to False
+
+
+@pytest.mark.parametrize("flag", [True, False])
+def test_checkpoint_flag_overrides_env(flag):
+    patched = dict(m.DEFAULT_CONFIG, checkpoint_enabled=not flag)
+    with mock.patch.object(m, "DEFAULT_CONFIG", patched):
+        cfg = m._build_run_config(SELECTIONS, checkpoint=flag)
+    assert cfg["checkpoint_enabled"] is flag

tests/test_cli_env_skip.py

@@ -82,5 +82,68 @@ class TestCliSkipsPromptsFromEnv(unittest.TestCase):
         self.assertEqual(sel["output_language"], "Japanese")
 
 
+@pytest.mark.unit
+class TestResearchDepthSkippedFromEnv(unittest.TestCase):
+    def test_both_round_envs_skip_depth_prompt(self):
+        import cli.main as m
+
+        env = {
+            "TRADINGAGENTS_MAX_DEBATE_ROUNDS": "2",
+            "TRADINGAGENTS_MAX_RISK_ROUNDS": "4",
+        }
+        fake_cfg = dict(m.DEFAULT_CONFIG)
+        fake_cfg.update({"max_debate_rounds": 2, "max_risk_discuss_rounds": 4})
+
+        with mock.patch.dict(os.environ, env, clear=False), \
+             mock.patch.object(m, "DEFAULT_CONFIG", fake_cfg), \
+             mock.patch.object(m, "fetch_announcements", return_value=None), \
+             mock.patch.object(m, "display_announcements"), \
+             mock.patch.object(m, "get_ticker", return_value="AAPL"), \
+             mock.patch.object(m, "get_analysis_date", return_value="2026-05-29"), \
+             mock.patch.object(m, "select_analysts", return_value=[]), \
+             mock.patch.object(m, "select_research_depth") as prompt_depth, \
+             mock.patch.object(m, "ensure_api_key"), \
+             mock.patch.object(m, "select_llm_provider", return_value=("openai", None)), \
+             mock.patch.object(m, "ask_output_language", return_value="English"), \
+             mock.patch.object(m, "select_shallow_thinking_agent", return_value="gpt-5.4-mini"), \
+             mock.patch.object(m, "select_deep_thinking_agent", return_value="gpt-5.5"), \
+             mock.patch.object(m, "ask_openai_reasoning_effort", return_value=None):
+            sel = m.get_user_selections()
+
+        # The research-depth prompt is skipped; the value comes from the env config.
+        prompt_depth.assert_not_called()
+        self.assertEqual(sel["research_depth"], 2)
+
+
+@pytest.mark.unit
+class TestReasoningEffortSkippedFromEnv(unittest.TestCase):
+    def test_effort_env_skips_step8_prompt(self):
+        import cli.main as m
+
+        env = {"TRADINGAGENTS_OPENAI_REASONING_EFFORT": "high"}
+        fake_cfg = dict(m.DEFAULT_CONFIG)
+        fake_cfg.update({"openai_reasoning_effort": "high"})
+
+        with mock.patch.dict(os.environ, env, clear=False), \
+             mock.patch.object(m, "DEFAULT_CONFIG", fake_cfg), \
+             mock.patch.object(m, "fetch_announcements", return_value=None), \
+             mock.patch.object(m, "display_announcements"), \
+             mock.patch.object(m, "get_ticker", return_value="AAPL"), \
+             mock.patch.object(m, "get_analysis_date", return_value="2026-05-29"), \
+             mock.patch.object(m, "select_analysts", return_value=[]), \
+             mock.patch.object(m, "select_research_depth", return_value=1), \
+             mock.patch.object(m, "ensure_api_key"), \
+             mock.patch.object(m, "select_llm_provider", return_value=("openai", None)), \
+             mock.patch.object(m, "ask_output_language", return_value="English"), \
+             mock.patch.object(m, "select_shallow_thinking_agent", return_value="gpt-5.4-mini"), \
+             mock.patch.object(m, "select_deep_thinking_agent", return_value="gpt-5.5"), \
+             mock.patch.object(m, "ask_openai_reasoning_effort") as prompt_effort:
+            sel = m.get_user_selections()
+
+        # The reasoning-effort prompt is skipped; the value comes from env config.
+        prompt_effort.assert_not_called()
+        self.assertEqual(sel["openai_reasoning_effort"], "high")
+
+
 if __name__ == "__main__":
     unittest.main()

tests/test_env_overrides.py

@@ -68,6 +68,27 @@ def test_bool_coercion(monkeypatch, raw, expected):
     assert dc.DEFAULT_CONFIG["checkpoint_enabled"] is expected
 
 
+def test_reasoning_thinking_overrides(monkeypatch):
+    """The provider reasoning/thinking knobs are env-configurable (non-interactive runs)."""
+    dc = _reload_with_env(
+        monkeypatch,
+        TRADINGAGENTS_OPENAI_REASONING_EFFORT="high",
+        TRADINGAGENTS_GOOGLE_THINKING_LEVEL="minimal",
+        TRADINGAGENTS_ANTHROPIC_EFFORT="low",
+    )
+    assert dc.DEFAULT_CONFIG["openai_reasoning_effort"] == "high"
+    assert dc.DEFAULT_CONFIG["google_thinking_level"] == "minimal"
+    assert dc.DEFAULT_CONFIG["anthropic_effort"] == "low"
+
+
+def test_reasoning_effort_defaults_to_none(monkeypatch):
+    """Unset reasoning/thinking knobs stay None so each provider uses its own default."""
+    dc = _reload_with_env(monkeypatch)
+    assert dc.DEFAULT_CONFIG["openai_reasoning_effort"] is None
+    assert dc.DEFAULT_CONFIG["google_thinking_level"] is None
+    assert dc.DEFAULT_CONFIG["anthropic_effort"] is None
+
+
 def test_empty_env_value_is_passthrough(monkeypatch):
     """Empty TRADINGAGENTS_* values must not clobber the built-in default."""
     dc = _reload_with_env(
@@ -82,13 +103,23 @@ def test_empty_env_value_is_passthrough(monkeypatch):
 def test_invalid_int_raises(monkeypatch):
     """Garbage int values should surface a ValueError at import, not silently misconfigure."""
     monkeypatch.setenv("TRADINGAGENTS_MAX_DEBATE_ROUNDS", "not-a-number")
-    with pytest.raises(ValueError):
+    with pytest.raises(ValueError, match="TRADINGAGENTS_MAX_DEBATE_ROUNDS"):
         importlib.reload(default_config_module)
     # Restore module state for subsequent tests in this process
     monkeypatch.delenv("TRADINGAGENTS_MAX_DEBATE_ROUNDS", raising=False)
     importlib.reload(default_config_module)
 
 
+@pytest.mark.parametrize("bad", ["treu", "flase", "maybe", "2", "enabled"])
+def test_invalid_bool_raises(monkeypatch, bad):
+    """A misspelled boolean must fail loudly (like ints) instead of silently False."""
+    monkeypatch.setenv("TRADINGAGENTS_CHECKPOINT_ENABLED", bad)
+    with pytest.raises(ValueError, match="TRADINGAGENTS_CHECKPOINT_ENABLED"):
+        importlib.reload(default_config_module)
+    monkeypatch.delenv("TRADINGAGENTS_CHECKPOINT_ENABLED", raising=False)
+    importlib.reload(default_config_module)
+
+
 def test_unknown_env_var_is_ignored(monkeypatch):
     """Env vars outside _ENV_OVERRIDES must not bleed into DEFAULT_CONFIG."""
     dc = _reload_with_env(

tests/test_openai_reasoning_effort.py

@@ -0,0 +1,42 @@
+"""OpenAI ``reasoning_effort`` is gated to reasoning models.
+
+Non-reasoning OpenAI models (gpt-4.1, gpt-4o, ...) 400 with "Unsupported
+parameter: 'reasoning.effort'". The client must drop the kwarg for those rather
+than forward it and crash the run. The GPT-5 family and the o-series accept it.
+"""
+
+import pytest
+
+from tradingagents.llm_clients.openai_client import (
+    OpenAIClient,
+    _supports_reasoning_effort,
+)
+
+
+@pytest.mark.parametrize(
+    "model,expected",
+    [
+        ("gpt-5.5", True), ("gpt-5.4", True), ("gpt-5.4-mini", True),
+        ("gpt-5.5-pro", True), ("o1", True), ("o3-mini", True),
+        ("gpt-4.1", False), ("gpt-4o", False), ("gpt-4o-mini", False),
+        ("gpt-3.5-turbo", False),
+    ],
+)
+def test_supports_reasoning_effort(model, expected):
+    assert _supports_reasoning_effort(model) is expected
+
+
+def _effort_on(model, monkeypatch):
+    # A fake key lets get_llm() construct the client without a network call.
+    monkeypatch.setenv("OPENAI_API_KEY", "test-key")
+    llm = OpenAIClient(model, provider="openai", reasoning_effort="low").get_llm()
+    return getattr(llm, "reasoning_effort", None)
+
+
+def test_reasoning_model_receives_effort(monkeypatch):
+    assert _effort_on("gpt-5.4-mini", monkeypatch) == "low"
+
+
+def test_non_reasoning_model_drops_effort(monkeypatch):
+    # gpt-4.1 would 400 with reasoning_effort — it must be dropped.
+    assert _effort_on("gpt-4.1", monkeypatch) is None

tradingagents/default_config.py

@@ -18,13 +18,35 @@ _ENV_OVERRIDES = {
     "TRADINGAGENTS_CHECKPOINT_ENABLED":   "checkpoint_enabled",
     "TRADINGAGENTS_BENCHMARK_TICKER":     "benchmark_ticker",
     "TRADINGAGENTS_TEMPERATURE":          "temperature",
+    # Provider-specific reasoning/thinking knobs (None = each provider's own
+    # default). Settable here for non-interactive runs; the CLI also offers an
+    # interactive choice, which is skipped when the matching var is set.
+    "TRADINGAGENTS_GOOGLE_THINKING_LEVEL":   "google_thinking_level",
+    "TRADINGAGENTS_OPENAI_REASONING_EFFORT": "openai_reasoning_effort",
+    "TRADINGAGENTS_ANTHROPIC_EFFORT":        "anthropic_effort",
 }
 
 
+_BOOL_TRUE = ("true", "1", "yes", "on")
+_BOOL_FALSE = ("false", "0", "no", "off")
+
+
 def _coerce(value: str, reference):
-    """Coerce env-var string to the type of the existing default value."""
+    """Coerce env-var string to the type of the existing default value.
+
+    Invalid values raise ``ValueError`` rather than silently falling back to a
+    default — a misspelled boolean (e.g. ``treu``) or non-numeric int should fail
+    loudly at startup, not quietly misconfigure an unattended run.
+    """
     if isinstance(reference, bool):
-        return value.strip().lower() in ("true", "1", "yes", "on")
+        normalized = value.strip().lower()
+        if normalized in _BOOL_TRUE:
+            return True
+        if normalized in _BOOL_FALSE:
+            return False
+        raise ValueError(
+            f"expected a boolean ({'/'.join(_BOOL_TRUE + _BOOL_FALSE)}), got {value!r}"
+        )
     if isinstance(reference, int) and not isinstance(reference, bool):
         return int(value)
     if isinstance(reference, float):
@@ -38,7 +60,10 @@ def _apply_env_overrides(config: dict) -> dict:
         raw = os.environ.get(env_var)
         if raw is None or raw == "":
             continue
-        config[key] = _coerce(raw, config.get(key))
+        try:
+            config[key] = _coerce(raw, config.get(key))
+        except ValueError as exc:
+            raise ValueError(f"Invalid value for {env_var}: {exc}") from exc
     return config
 
 

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)