feat(endpoints): support python -m litellm/copilot modules and literal model lists for list_models_cmd

Author: jzhu

code_assistant_manager/endpoints.py

@@ -267,27 +267,65 @@ def fetch_models(
         try:
             toolbox_dir = Path(__file__).parent.parent
             # Security: Use shlex.split() instead of shell=True to prevent injection
+            import contextlib
+            import importlib
+            import io
             import shlex
+            import shutil
 
-            result = subprocess.run(
-                shlex.split(list_cmd),
-                shell=False,
-                capture_output=True,
-                text=True,
-                cwd=toolbox_dir,
-                env=env,
-                timeout=60,
-            )
+            tokens = shlex.split(list_cmd)
+
+            output = ""
+
+            # Support running specific python -m modules by calling them directly
+            if (
+                len(tokens) >= 3
+                and tokens[1] == "-m"
+                and tokens[2]
+                in (
+                    "code_assistant_manager.litellm_models",
+                    "code_assistant_manager.copilot_models",
+                )
+            ):
+                module_name = tokens[2]
+                try:
+                    mod = importlib.import_module(module_name)
+                    buf = io.StringIO()
+                    with contextlib.redirect_stdout(buf):
+                        # Call list_models() which prints model IDs
+                        mod.list_models()
+                    output = buf.getvalue().strip()
+                except Exception as e:
+                    print(f"Warning: Failed to load module {module_name}: {e}")
+                    return True, models
+
+            # If first token is not an executable on PATH, treat the value as literal model list
+            elif tokens and shutil.which(tokens[0]) is None:
+                output = " ".join(tokens).strip()
+
+            else:
+                # Fallback: run as subprocess
+                result = subprocess.run(
+                    tokens,
+                    shell=False,
+                    capture_output=True,
+                    text=True,
+                    cwd=toolbox_dir,
+                    env=env,
+                    timeout=60,
+                )
 
-            output = result.stdout.strip()
-            if result.returncode != 0:
-                print(f"Warning: Command failed with return code {result.returncode}")
-                if result.stderr:
-                    print(f"Command stderr: {result.stderr}")
-                return True, models
-            if not output:
-                print("Warning: Command returned no output")
-                return True, models
+                output = result.stdout.strip()
+                if result.returncode != 0:
+                    print(
+                        f"Warning: Command failed with return code {result.returncode}"
+                    )
+                    if result.stderr:
+                        print(f"Command stderr: {result.stderr}")
+                    return True, models
+                if not output:
+                    print("Warning: Command returned no output")
+                    return True, models
 
             # Parse output
             models = self._parse_models_output(output)