fix(anthropic): use model-native output limits instead of hardcoded 16K (#3426)

The Anthropic adapter defaulted to max_tokens=16384 when no explicit value
was configured.  This severely limits thinking-enabled models where thinking
tokens count toward max_tokens:

- Claude Opus 4.6 supports 128K output but was capped at 16K
- Claude Sonnet 4.6 supports 64K output but was capped at 16K

With extended thinking (adaptive or budget-based), the model could exhaust
the entire 16K on reasoning, leaving zero tokens for the actual response.
This caused two user-visible errors:
- 'Response truncated (finish_reason=length)' — thinking consumed most tokens
- 'Response only contains think block with no content' — thinking consumed all

Fix: add _ANTHROPIC_OUTPUT_LIMITS lookup table (sourced from Anthropic docs
and Cline's model catalog) and use the model's actual output limit as the
default.  Unknown future models default to 128K (the current maximum).

Also adds context_length clamping: if the user configured a smaller context
window (e.g. custom endpoint), max_tokens is clamped to context_length - 1
to avoid exceeding the window.

Closes #2706

agent/anthropic_adapter.py

@@ -35,6 +35,54 @@ ADAPTIVE_EFFORT_MAP = {
     "minimal": "low",
 }
 
+# ── Max output token limits per Anthropic model ───────────────────────
+# Source: Anthropic docs + Cline model catalog.  Anthropic's API requires
+# max_tokens as a mandatory field.  Previously we hardcoded 16384, which
+# starves thinking-enabled models (thinking tokens count toward the limit).
+_ANTHROPIC_OUTPUT_LIMITS = {
+    # Claude 4.6
+    "claude-opus-4-6":   128_000,
+    "claude-sonnet-4-6":  64_000,
+    # Claude 4.5
+    "claude-opus-4-5":    64_000,
+    "claude-sonnet-4-5":  64_000,
+    "claude-haiku-4-5":   64_000,
+    # Claude 4
+    "claude-opus-4":      32_000,
+    "claude-sonnet-4":    64_000,
+    # Claude 3.7
+    "claude-3-7-sonnet": 128_000,
+    # Claude 3.5
+    "claude-3-5-sonnet":   8_192,
+    "claude-3-5-haiku":    8_192,
+    # Claude 3
+    "claude-3-opus":       4_096,
+    "claude-3-sonnet":     4_096,
+    "claude-3-haiku":      4_096,
+}
+
+# For any model not in the table, assume the highest current limit.
+# Future Anthropic models are unlikely to have *less* output capacity.
+_ANTHROPIC_DEFAULT_OUTPUT_LIMIT = 128_000
+
+
+def _get_anthropic_max_output(model: str) -> int:
+    """Look up the max output token limit for an Anthropic model.
+
+    Uses substring matching against _ANTHROPIC_OUTPUT_LIMITS so date-stamped
+    model IDs (claude-sonnet-4-5-20250929) and variant suffixes (:1m, :fast)
+    resolve correctly.  Longest-prefix match wins to avoid e.g. "claude-3-5"
+    matching before "claude-3-5-sonnet".
+    """
+    m = model.lower()
+    best_key = ""
+    best_val = _ANTHROPIC_DEFAULT_OUTPUT_LIMIT
+    for key, val in _ANTHROPIC_OUTPUT_LIMITS.items():
+        if key in m and len(key) > len(best_key):
+            best_key = key
+            best_val = val
+    return best_val
+
 
 def _supports_adaptive_thinking(model: str) -> bool:
     """Return True for Claude 4.6 models that support adaptive thinking."""
@@ -818,9 +866,15 @@ def build_anthropic_kwargs(
     tool_choice: Optional[str] = None,
     is_oauth: bool = False,
     preserve_dots: bool = False,
+    context_length: Optional[int] = None,
 ) -> Dict[str, Any]:
     """Build kwargs for anthropic.messages.create().
 
+    When *max_tokens* is None, the model's native output limit is used
+    (e.g. 128K for Opus 4.6, 64K for Sonnet 4.6).  If *context_length*
+    is provided, the effective limit is clamped so it doesn't exceed
+    the context window.
+
     When *is_oauth* is True, applies Claude Code compatibility transforms:
     system prompt prefix, tool name prefixing, and prompt sanitization.
 
@@ -831,7 +885,12 @@ def build_anthropic_kwargs(
     anthropic_tools = convert_tools_to_anthropic(tools) if tools else []
 
     model = normalize_model_name(model, preserve_dots=preserve_dots)
-    effective_max_tokens = max_tokens or 16384
+    effective_max_tokens = max_tokens or _get_anthropic_max_output(model)
+
+    # Clamp to context window if the user set a lower context_length
+    # (e.g. custom endpoint with limited capacity).
+    if context_length and effective_max_tokens > context_length:
+        effective_max_tokens = max(context_length - 1, 1)
 
     # ── OAuth: Claude Code identity ──────────────────────────────────
     if is_oauth: