fix: parse OpenRouter/Nous "in the output" error format in parse_available_output_tokens_from_error()

agent/model_metadata.py

@@ -957,9 +957,18 @@ def parse_available_output_tokens_from_error(error_msg: str) -> Optional[int]:
     error_lower = error_msg.lower()
 
     # Must look like an output-cap error, not a prompt-length error.
+    # OpenRouter/Nous Research format (OpenAI-compatible):
+    #   "5683 of text input, 13410 of tool input, 262000 in the output"
+    # Note: these errors do NOT contain the word "max_tokens" — they use
+    # "in the output" to describe the output cap.
     is_output_cap_error = (
+        re.search(r'\d+\s+in\s+the\s+output', error_lower) is not None
+    ) or (
         "max_tokens" in error_lower
-        and ("available_tokens" in error_lower or "available tokens" in error_lower)
+        and (
+            "available_tokens" in error_lower
+            or "available tokens" in error_lower
+        )
     )
     if not is_output_cap_error:
         return None
@@ -972,6 +981,23 @@ def parse_available_output_tokens_from_error(error_msg: str) -> Optional[int]:
         # fallback: last number after "=" in expressions like "200000 - 190000 = 10000"
         r'=\s*(\d+)\s*$',
     ]
+
+    # OpenRouter/Nous format: compute available = context_length - text_input - tool_input
+    #   "maximum context length is 256000 ... 5683 of text input ... 13410 of tool input"
+    or_match = re.search(
+        r'maximum\s+context\s+length\s+is\s+(\d+).*?'
+        r'(\d+)\s+of\s+text\s+input.*?'
+        r'(\d+)\s+of\s+tool\s+input',
+        error_lower
+    )
+    if or_match:
+        max_ctx = int(or_match.group(1))
+        text_input = int(or_match.group(2))
+        tool_input = int(or_match.group(3))
+        available = max_ctx - text_input - tool_input
+        if available >= 1:
+            return available
+
     for pattern in patterns:
         match = re.search(pattern, error_lower)
         if match:

tests/agent/test_model_metadata.py

@@ -1168,6 +1168,64 @@ def test_number_outside_reasonable_range(self):
         assert parse_context_limit_from_error(msg) is None
 
 
+class TestParseAvailableOutputTokensFromError:
+    """Tests for parse_available_output_tokens_from_error() — detects "output cap
+    too large" errors and returns how many output tokens would fit."""
+
+    # ── Anthropic format ──────────────────────────────────────────────
+
+    def test_anthropic_available_tokens(self):
+        msg = (
+            "max_tokens: 32768 > context_window: 200000 - input_tokens: 190000"
+            " = available_tokens: 10000"
+        )
+        assert parse_available_output_tokens_from_error(msg) == 10000
+
+    def test_anthropic_available_tokens_with_space(self):
+        msg = "max_tokens: 50000 exceeds available tokens: 30000"
+        assert parse_available_output_tokens_from_error(msg) == 30000
+
+    # ── OpenRouter / Nous Research format ─────────────────────────────
+
+    def test_openrouter_format_returns_calculated_available(self):
+        """Real Nous Research error — compute available = context - text - tool."""
+        msg = (
+            "This endpoint's maximum context length is 256000 tokens."
+            " However, you requested about 281093 tokens"
+            " (5683 of text input, 13410 of tool input, 262000 in the output)."
+            " Please reduce the length of either one, or use the"
+            " context-compression plugin to compress your prompt automatically."
+        )
+        # available = 256000 - 5683 - 13410 = 236907
+        assert parse_available_output_tokens_from_error(msg) == 236907
+
+    def test_openrouter_large_tool_input(self):
+        """Scenario with 50K tool schemas."""
+        msg = (
+            "maximum context length is 131072 tokens."
+            " However, you requested about 200000 tokens"
+            " (10000 of text input, 50000 of tool input, 140000 in the output)."
+        )
+        # available = 131072 - 10000 - 50000 = 71072
+        assert parse_available_output_tokens_from_error(msg) == 71072
+
+    # ── Negative cases ────────────────────────────────────────────────
+
+    def test_plain_context_overflow_not_output_cap(self):
+        """A prompt-too-long error without max_tokens mention."""
+        msg = (
+            "This model's maximum context length is 32768 tokens."
+            " However, your messages resulted in 45000 tokens."
+        )
+        assert parse_available_output_tokens_from_error(msg) is None
+
+    def test_completely_unrelated_error(self):
+        assert parse_available_output_tokens_from_error("Invalid API key") is None
+
+    def test_empty_string(self):
+        assert parse_available_output_tokens_from_error("") is None
+
+
 # =========================================================================
 # Persistent context length cache
 # =========================================================================