diff --git a/agent/anthropic_adapter.py b/agent/anthropic_adapter.py
index 5e36b1f37..0a04bee01 100644
--- a/agent/anthropic_adapter.py
+++ b/agent/anthropic_adapter.py
@@ -117,6 +117,63 @@ def _get_anthropic_max_output(model: str) -> int:
     return best_val
 
 
+def _resolve_positive_anthropic_max_tokens(value) -> Optional[int]:
+    """Return ``value`` floored to a positive int, or ``None`` if it is not a
+    finite positive number. Ported from openclaw/openclaw#66664.
+
+    Anthropic's Messages API rejects ``max_tokens`` values that are 0,
+    negative, non-integer, or non-finite with HTTP 400. Python's ``or``
+    idiom (``max_tokens or fallback``) correctly catches ``0`` but lets
+    negative ints and fractional floats (``-1``, ``0.5``) through to the
+    API, producing a user-visible failure instead of a local error.
+    """
+    # Booleans are a subclass of int — exclude explicitly so ``True`` doesn't
+    # silently become 1 and ``False`` doesn't become 0.
+    if isinstance(value, bool):
+        return None
+    if not isinstance(value, (int, float)):
+        return None
+    try:
+        import math
+        if not math.isfinite(value):
+            return None
+    except Exception:
+        return None
+    floored = int(value)  # truncates toward zero for floats
+    return floored if floored > 0 else None
+
+
+def _resolve_anthropic_messages_max_tokens(
+    requested,
+    model: str,
+    context_length: Optional[int] = None,
+) -> int:
+    """Resolve the ``max_tokens`` budget for an Anthropic Messages call.
+
+    Prefers ``requested`` when it is a positive finite number; otherwise
+    falls back to the model's output ceiling. Raises ``ValueError`` if no
+    positive budget can be resolved (should not happen with current model
+    table defaults, but guards against a future regression where
+    ``_get_anthropic_max_output`` could return ``0``).
+
+    Separately, callers apply a context-window clamp — this resolver does
+    not, to keep the positive-value contract independent of endpoint
+    specifics.
+
+    Ported from openclaw/openclaw#66664 (resolveAnthropicMessagesMaxTokens).
+    """
+    resolved = _resolve_positive_anthropic_max_tokens(requested)
+    if resolved is not None:
+        return resolved
+    fallback = _get_anthropic_max_output(model)
+    if fallback > 0:
+        return fallback
+    raise ValueError(
+        f"Anthropic Messages adapter requires a positive max_tokens value for "
+        f"model {model!r}; got {requested!r} and no model default resolved."
+    )
+
+
 def _supports_adaptive_thinking(model: str) -> bool:
     """Return True for Claude 4.6+ models that support adaptive thinking."""
     return any(v in model for v in _ADAPTIVE_THINKING_SUBSTRINGS)
@@ -1391,7 +1448,12 @@ def build_anthropic_kwargs(
 
     model = normalize_model_name(model, preserve_dots=preserve_dots)
     # effective_max_tokens = output cap for this call (≠ total context window)
-    effective_max_tokens = max_tokens or _get_anthropic_max_output(model)
+    # Use the resolver helper so non-positive values (negative ints,
+    # fractional floats, NaN, non-numeric) fail locally with a clear error
+    # rather than 400-ing at the Anthropic API. See openclaw/openclaw#66664.
+    effective_max_tokens = _resolve_anthropic_messages_max_tokens(
+        max_tokens, model, context_length=context_length
+    )
 
     # Clamp output cap to fit inside the total context window.
     # Only matters for small custom endpoints where context_length < native
diff --git a/tests/agent/test_anthropic_adapter.py b/tests/agent/test_anthropic_adapter.py
index b947a2df8..dedf3e125 100644
--- a/tests/agent/test_anthropic_adapter.py
+++ b/tests/agent/test_anthropic_adapter.py
@@ -1659,3 +1659,91 @@ class TestToolChoice:
             tool_choice="search",
         )
         assert kwargs["tool_choice"] == {"type": "tool", "name": "search"}
+
+
+
+# ---------------------------------------------------------------------------
+# max_tokens resolver — openclaw/openclaw#66664 port
+# ---------------------------------------------------------------------------
+
+from agent.anthropic_adapter import (
+    _resolve_positive_anthropic_max_tokens,
+    _resolve_anthropic_messages_max_tokens,
+)
+
+
+class TestResolvePositiveMaxTokens:
+    """Unit tests for the positive-int resolver helper."""
+
+    def test_positive_int_passes_through(self):
+        assert _resolve_positive_anthropic_max_tokens(8192) == 8192
+
+    def test_zero_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(0) is None
+
+    def test_negative_int_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(-1) is None
+        assert _resolve_positive_anthropic_max_tokens(-500) is None
+
+    def test_fractional_float_floored_and_kept_if_positive(self):
+        # 8192.7 -> 8192, still positive
+        assert _resolve_positive_anthropic_max_tokens(8192.7) == 8192
+
+    def test_small_positive_float_below_one_returns_none(self):
+        # 0.5 floors to 0, which is not positive
+        assert _resolve_positive_anthropic_max_tokens(0.5) is None
+
+    def test_negative_float_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(-1.5) is None
+
+    def test_nan_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(float("nan")) is None
+
+    def test_infinity_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(float("inf")) is None
+        assert _resolve_positive_anthropic_max_tokens(float("-inf")) is None
+
+    def test_bool_true_returns_none(self):
+        # True is an int subclass but semantically never a real max_tokens value
+        assert _resolve_positive_anthropic_max_tokens(True) is None
+        assert _resolve_positive_anthropic_max_tokens(False) is None
+
+    def test_string_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens("8192") is None
+
+    def test_none_returns_none(self):
+        assert _resolve_positive_anthropic_max_tokens(None) is None
+
+
+class TestResolveMessagesMaxTokens:
+    """Integration tests for the full Messages resolver."""
+
+    def test_positive_requested_wins(self):
+        assert _resolve_anthropic_messages_max_tokens(
+            8192, "claude-opus-4-6"
+        ) == 8192
+
+    def test_zero_falls_back_to_model_default(self):
+        # Should use _get_anthropic_max_output(model), not crash
+        result = _resolve_anthropic_messages_max_tokens(0, "claude-opus-4-6")
+        assert result > 0
+
+    def test_none_falls_back_to_model_default(self):
+        result = _resolve_anthropic_messages_max_tokens(None, "claude-opus-4-6")
+        assert result > 0
+
+    def test_negative_falls_back_to_model_default(self):
+        # Previously leaked -1 to the API; now falls back safely
+        result = _resolve_anthropic_messages_max_tokens(-1, "claude-opus-4-6")
+        assert result > 0
+
+    def test_fractional_positive_floored(self):
+        assert _resolve_anthropic_messages_max_tokens(
+            8192.5, "claude-opus-4-6"
+        ) == 8192
+
+    def test_sub_one_float_falls_back(self):
+        # 0.5 floors to 0 -> not positive -> falls back to model ceiling
+        result = _resolve_anthropic_messages_max_tokens(0.5, "claude-opus-4-6")
+        assert result > 0
+        assert result != 0