Fixes issues when running Goose on Windows.

src/metacoder/coders/base_coder.py

@@ -173,11 +173,15 @@ def run_process(
         """
         if env is None:
             env = self.expand_env(self.env)
+
+        # Decode the child process output as UTF-8 (instead of default encoding)
         process = subprocess.Popen(
             command,
             stdout=subprocess.PIPE,
             stderr=subprocess.PIPE,
             text=True,
+            encoding="utf-8",
+            errors="replace",  # avoid crashes on the occasional bad byte
             env=env,
             bufsize=1,
             universal_newlines=True,
@@ -189,6 +193,15 @@ def run_process(
         # check verbosity level
         quiet_mode = logger.getEffectiveLevel() <= logging.INFO
 
+        # Ensure our own stdout/stderr won't choke on non-ASCII (Windows consoles often do).
+        for s in (sys.stdout, sys.stderr):
+            try:
+                s.reconfigure(encoding="utf-8", errors="replace")  # Python 3.7+
+            except Exception as e:
+                logger.info(f"{e}")
+                pass  # OK if not available (e.g., redirected or older Python)
+
+        # lines are already str decoded as UTF-8
         def stream_output(pipe, output_lines, stream):
             for line in iter(pipe.readline, ""):
                 if not quiet_mode:

src/metacoder/coders/claude.py

@@ -246,7 +246,7 @@ def parse_jsonl_line(text: str) -> dict[str, Any]:
                 ao.tool_uses = tool_uses
 
             end_time = time.time()
-            logger.info(f"🤖 Command took {end_time - start_time} seconds")
+            logger.info(f"🤖 Command took {end_time - start_time:.2f} seconds")
             ao.total_cost_usd = total_cost_usd
             ao.success = not is_error
             if not ao.success:

src/metacoder/coders/codex.py

@@ -115,7 +115,7 @@ def run(self, input_text: str) -> CoderOutput:
             if "result" in message:
                 ao.result_text = message["result"]
         end_time = time.time()
-        print(f"🤖 Command took {end_time - start_time} seconds")
+        print(f"🤖 Command took {end_time - start_time:.2f} seconds")
         ao.total_cost_usd = total_cost_usd
         ao.success = not is_error
         if not ao.success:

src/metacoder/coders/gemini.py

@@ -156,7 +156,7 @@ def run(self, input_text: str) -> CoderOutput:
                 )
 
             end_time = time.time()
-            logger.info(f"💎 Command took {end_time - start_time} seconds")
+            logger.info(f"💎 Command took {end_time - start_time:.2f} seconds")
 
             # Parse the output
             ao = CoderOutput(stdout=result.stdout, stderr=result.stderr)

src/metacoder/coders/goose.py

@@ -156,7 +156,7 @@ def run(self, input_text: str) -> CoderOutput:
             result = self.run_process(command, env)
             end_time = time.time()
             ao = CoderOutput(stdout=result.stdout, stderr=result.stderr)
-            logger.info(f"🦆 Command took {end_time - start_time} seconds")
+            logger.info(f"🦆 Command took {end_time - start_time:.2f} seconds")
             # look in output text for a file like: logging to ./.local/share/goose/sessions/20250613_120403.jsonl
             session_file: Optional[Path] = None
             for line in result.stdout.split("\n"):
@@ -165,7 +165,7 @@ def run(self, input_text: str) -> CoderOutput:
                     session_file = Path(session_file_str)
                     break
             if session_file and session_file.exists():
-                with open(session_file, "r") as f:
+                with open(session_file, "r", encoding="utf-8") as f:
                     ao.structured_messages = [
                         json.loads(line) for line in f if line.strip()
                     ]

src/metacoder/coders/qwen.py

@@ -90,7 +90,7 @@ def run(self, input_text: str) -> CoderOutput:
                 )
 
             end_time = time.time()
-            print(f"🤖 Command took {end_time - start_time} seconds")
+            print(f"🤖 Command took {end_time - start_time:.2f} seconds")
 
             # Create output - Qwen CLI doesn't provide structured output
             ao = CoderOutput(

src/metacoder/evals/judges.py

@@ -0,0 +1,93 @@
+# metacoder/evals/judges.py
+import logging
+import os
+
+from anthropic import Anthropic
+from anthropic.types import MessageParam, TextBlockParam, TextBlock
+
+from deepeval.models.base_model import DeepEvalBaseLLM
+
+logger = logging.getLogger(__name__)
+
+
+class ClaudeJudge(DeepEvalBaseLLM):
+    """
+    Wraps Anthropic's Claude models so they can be used as
+    the `model` parameter to DeepEval metrics like GEval.
+    """
+
+    # Note: Anthropic models can be listed via:
+    # curl https://api.anthropic.com/v1/models --header "x-api-key: %ANTHROPIC_API_KEY%" --header "anthropic-version: 2023-06-01"
+    # {"data": [{"type": "model", "id": "claude-opus-4-1-20250805", "display_name": "Claude Opus 4.1", "created_at": "2025-08-05T00:00:00Z"}, ... ]}
+    # Current list (September 3, 2025):
+    # claude-opus-4-1-20250805, claude-opus-4-20250514, claude-sonnet-4-20250514, claude-3-7-sonnet-20250219,
+    # claude-3-5-sonnet-20241022, claude-3-5-haiku-20241022, claude-3-5-sonnet-20240620, claude-3-haiku-20240307,
+    # claude-3-opus-20240229
+
+    def __init__(
+        self,
+        model_name: str = "claude-sonnet-4-20250514",
+        max_tokens: int = 1024,
+        temperature: float = 0.0,
+    ):
+        super().__init__()
+        api_key = os.getenv("ANTHROPIC_API_KEY")
+        if not api_key:
+            raise Exception("ANTHROPIC_API_KEY is not set in environment")
+        self.client = Anthropic(api_key=api_key)
+        self.model_name = model_name
+        self.max_tokens = max_tokens
+        self.temperature = temperature
+
+    def load_model(self):
+        return self
+
+    def generate(self, prompt: str) -> str:
+        # Build typed content blocks and messages to satisfy the SDK's type hints
+        content: list[TextBlockParam] = [{"type": "text", "text": prompt}]
+        messages: list[MessageParam] = [{"role": "user", "content": content}]
+        resp = self.client.messages.create(
+            model=self.model_name,
+            max_tokens=self.max_tokens,
+            temperature=self.temperature,
+            messages=messages,
+        )
+        # anthropic returns a list of content blocks; collect only the text blocks.
+        parts: list[str] = []
+        for block in resp.content:
+            if isinstance(block, TextBlock):
+                parts.append(block.text)
+        return "".join(parts)
+
+    async def a_generate(self, prompt: str) -> str:
+        # for now just call the sync path
+        return self.generate(prompt)
+
+    def get_model_name(self) -> str:
+        return self.model_name
+
+    def has_available_quota(self) -> bool:
+        """
+        Try a very lightweight request to check if quota is available.
+        Returns True if quota exists, False if Anthropic responds with
+        quota-related errors.
+        """
+        try:
+            # Use a minimal "ping" request
+            content: list[TextBlockParam] = [{"type": "text", "text": "ping"}]
+            messages: list[MessageParam] = [{"role": "user", "content": content}]
+            self.client.messages.create(
+                model=self.model_name,
+                max_tokens=1,  # cheapest possible
+                temperature=0.0,
+                messages=messages,
+            )
+            return True
+        except Exception as e:
+            msg = str(e).lower()
+            # Check for insufficient quota:
+            # 400 Bad Request. Message: Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits.
+            if "credit balance is too low" in msg or "400" in msg:
+                logger.warning(f"ClaudeJudge quota check failed: {e}")
+                return False
+            raise