code-puppy 0.0.348__py3-none-any.whl → 0.0.361__py3-none-any.whl

code_puppy/claude_cache_client.py

@@ -5,6 +5,11 @@ ClaudeCacheAsyncClient: httpx client that tries to patch /v1/messages bodies.
 We now also expose `patch_anthropic_client_messages` which monkey-patches
 AsyncAnthropic.messages.create() so we can inject cache_control BEFORE
 serialization, avoiding httpx/Pydantic internals.
+
+This module also handles:
+- Tool name prefixing/unprefixing for Claude Code OAuth compatibility
+- Header transformations (anthropic-beta, user-agent)
+- URL modifications (adding ?beta=true query param)
 """
 
 from __future__ import annotations
@@ -12,8 +17,10 @@ from __future__ import annotations
 import base64
 import json
 import logging
+import re
 import time
 from typing import Any, Callable, MutableMapping
+from urllib.parse import parse_qs, urlencode, urlparse, urlunparse
 
 import httpx
 
@@ -22,6 +29,13 @@ logger = logging.getLogger(__name__)
 # Refresh token if it's older than 1 hour (3600 seconds)
 TOKEN_MAX_AGE_SECONDS = 3600
 
+# Tool name prefix for Claude Code OAuth compatibility
+# Tools are prefixed on outgoing requests and unprefixed on incoming responses
+TOOL_PREFIX = "cp_"
+
+# User-Agent to send with Claude Code OAuth requests
+CLAUDE_CLI_USER_AGENT = "claude-cli/2.1.2 (external, cli)"
+
 try:
     from anthropic import AsyncAnthropic
 except ImportError:  # pragma: no cover - optional dep
@@ -29,6 +43,22 @@ except ImportError:  # pragma: no cover - optional dep
 
 
 class ClaudeCacheAsyncClient(httpx.AsyncClient):
+    """Async HTTP client with Claude Code OAuth transformations.
+
+    Handles:
+    - Cache control injection for prompt caching
+    - Tool name prefixing on outgoing requests
+    - Tool name unprefixing on incoming streaming responses
+    - Header transformations (anthropic-beta, user-agent)
+    - URL modifications (adding ?beta=true)
+    - Proactive token refresh
+    """
+
+    # Regex pattern for unprefixing tool names in streaming responses
+    _TOOL_UNPREFIX_PATTERN = re.compile(
+        rf'"name"\s*:\s*"{re.escape(TOOL_PREFIX)}([^"]+)"'
+    )
+
     def _get_jwt_age_seconds(self, token: str | None) -> float | None:
         """Decode a JWT and return its age in seconds.
 
@@ -107,9 +137,100 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
             )
         return should_refresh
 
+    @staticmethod
+    def _prefix_tool_names(body: bytes) -> bytes | None:
+        """Prefix all tool names in the request body with TOOL_PREFIX.
+
+        This is required for Claude Code OAuth compatibility - tools must be
+        prefixed on outgoing requests and unprefixed on incoming responses.
+        """
+        try:
+            data = json.loads(body.decode("utf-8"))
+        except Exception:
+            return None
+
+        if not isinstance(data, dict):
+            return None
+
+        tools = data.get("tools")
+        if not isinstance(tools, list) or not tools:
+            return None
+
+        modified = False
+        for tool in tools:
+            if isinstance(tool, dict) and "name" in tool:
+                name = tool["name"]
+                if name and not name.startswith(TOOL_PREFIX):
+                    tool["name"] = f"{TOOL_PREFIX}{name}"
+                    modified = True
+
+        if not modified:
+            return None
+
+        return json.dumps(data).encode("utf-8")
+
+    def _unprefix_tool_names_in_text(self, text: str) -> str:
+        """Remove TOOL_PREFIX from tool names in streaming response text."""
+        return self._TOOL_UNPREFIX_PATTERN.sub(r'"name": "\1"', text)
+
+    @staticmethod
+    def _transform_headers_for_claude_code(
+        headers: MutableMapping[str, str],
+    ) -> None:
+        """Transform headers for Claude Code OAuth compatibility.
+
+        - Sets user-agent to claude-cli
+        - Merges anthropic-beta headers appropriately
+        - Removes x-api-key (using Bearer auth instead)
+        """
+        # Set user-agent
+        headers["user-agent"] = CLAUDE_CLI_USER_AGENT
+
+        # Handle anthropic-beta header
+        incoming_beta = headers.get("anthropic-beta", "")
+        incoming_betas = [b.strip() for b in incoming_beta.split(",") if b.strip()]
+
+        # Check if claude-code beta was explicitly requested
+        include_claude_code = "claude-code-20250219" in incoming_betas
+
+        # Build merged betas list
+        merged_betas = [
+            "oauth-2025-04-20",
+            "interleaved-thinking-2025-05-14",
+        ]
+        if include_claude_code:
+            merged_betas.append("claude-code-20250219")
+
+        headers["anthropic-beta"] = ",".join(merged_betas)
+
+        # Remove x-api-key if present (we use Bearer auth)
+        for key in ["x-api-key", "X-API-Key", "X-Api-Key"]:
+            if key in headers:
+                del headers[key]
+
+    @staticmethod
+    def _add_beta_query_param(url: httpx.URL) -> httpx.URL:
+        """Add ?beta=true query parameter to the URL if not already present."""
+        # Parse the URL
+        parsed = urlparse(str(url))
+        query_params = parse_qs(parsed.query)
+
+        # Only add if not already present
+        if "beta" not in query_params:
+            query_params["beta"] = ["true"]
+            # Rebuild query string
+            new_query = urlencode(query_params, doseq=True)
+            # Rebuild URL
+            new_parsed = parsed._replace(query=new_query)
+            return httpx.URL(urlunparse(new_parsed))
+
+        return url
+
     async def send(
         self, request: httpx.Request, *args: Any, **kwargs: Any
     ) -> httpx.Response:  # type: ignore[override]
+        is_messages_endpoint = request.url.path.endswith("/v1/messages")
+
         # Proactive token refresh: check JWT age before every request
         if not request.extensions.get("claude_oauth_refresh_attempted"):
             try:
@@ -131,50 +252,88 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
             except Exception as exc:
                 logger.debug("Error during proactive token refresh check: %s", exc)
 
-        try:
-            if request.url.path.endswith("/v1/messages"):
+        # Apply Claude Code OAuth transformations for /v1/messages
+        if is_messages_endpoint:
+            try:
                 body_bytes = self._extract_body_bytes(request)
+                headers = dict(request.headers)
+                url = request.url
+                body_modified = False
+                headers_modified = False
+
+                # 1. Transform headers for Claude Code OAuth
+                self._transform_headers_for_claude_code(headers)
+                headers_modified = True
+
+                # 2. Add ?beta=true query param
+                url = self._add_beta_query_param(url)
+
+                # 3. Prefix tool names in request body
                 if body_bytes:
-                    updated = self._inject_cache_control(body_bytes)
-                    if updated is not None:
-                        # Rebuild a request with the updated body and transplant internals
-                        try:
-                            rebuilt = self.build_request(
-                                method=request.method,
-                                url=request.url,
-                                headers=request.headers,
-                                content=updated,
-                            )
-
-                            # Copy core internals so httpx uses the modified body/stream
-                            if hasattr(rebuilt, "_content"):
-                                setattr(request, "_content", rebuilt._content)  # type: ignore[attr-defined]
-                            if hasattr(rebuilt, "stream"):
-                                request.stream = rebuilt.stream
-                            if hasattr(rebuilt, "extensions"):
-                                request.extensions = rebuilt.extensions
-
-                            # Ensure Content-Length matches the new body
-                            request.headers["Content-Length"] = str(len(updated))
-
-                        except Exception:
-                            # Swallow instrumentation errors; do not break real calls.
-                            pass
-        except Exception:
-            # Swallow wrapper errors; do not break real calls.
-            pass
+                    prefixed_body = self._prefix_tool_names(body_bytes)
+                    if prefixed_body is not None:
+                        body_bytes = prefixed_body
+                        body_modified = True
+
+                    # 4. Inject cache_control
+                    cached_body = self._inject_cache_control(body_bytes)
+                    if cached_body is not None:
+                        body_bytes = cached_body
+                        body_modified = True
+
+                # Rebuild request if anything changed
+                if body_modified or headers_modified or url != request.url:
+                    try:
+                        rebuilt = self.build_request(
+                            method=request.method,
+                            url=url,
+                            headers=headers,
+                            content=body_bytes,
+                        )
+
+                        # Copy core internals so httpx uses the modified body/stream
+                        if hasattr(rebuilt, "_content"):
+                            setattr(request, "_content", rebuilt._content)  # type: ignore[attr-defined]
+                        if hasattr(rebuilt, "stream"):
+                            request.stream = rebuilt.stream
+                        if hasattr(rebuilt, "extensions"):
+                            request.extensions = rebuilt.extensions
+
+                        # Update URL
+                        request.url = url
+
+                        # Update headers
+                        for key, value in headers.items():
+                            request.headers[key] = value
+
+                        # Ensure Content-Length matches the new body
+                        if body_bytes:
+                            request.headers["Content-Length"] = str(len(body_bytes))
+
+                    except Exception as exc:
+                        logger.debug("Error rebuilding request: %s", exc)
+
+            except Exception as exc:
+                logger.debug("Error in Claude Code transformations: %s", exc)
+
+        # Send the request
         response = await super().send(request, *args, **kwargs)
+
+        # Transform streaming response to unprefix tool names
+        if is_messages_endpoint and response.status_code == 200:
+            try:
+                response = self._wrap_response_with_tool_unprefixing(response, request)
+            except Exception as exc:
+                logger.debug("Error wrapping response for tool unprefixing: %s", exc)
+
+        # Handle auth errors with token refresh
         try:
-            # Check for both 401 and 400 - Anthropic/Cloudflare may return 400 for auth errors
-            # Also check if it's a Cloudflare HTML error response
             if response.status_code in (400, 401) and not request.extensions.get(
                 "claude_oauth_refresh_attempted"
             ):
-                # Determine if this is an auth error (including Cloudflare HTML errors)
                 is_auth_error = response.status_code == 401
 
                 if response.status_code == 400:
-                    # Check if this is a Cloudflare HTML error
                     is_auth_error = self._is_cloudflare_html_error(response)
                     if is_auth_error:
                         logger.info(
@@ -203,8 +362,64 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
                         logger.warning("Token refresh failed, returning original error")
         except Exception as exc:
             logger.debug("Error during token refresh attempt: %s", exc)
+
         return response
 
+    def _wrap_response_with_tool_unprefixing(
+        self, response: httpx.Response, request: httpx.Request
+    ) -> httpx.Response:
+        """Wrap a streaming response to unprefix tool names.
+
+        Creates a new response with a transformed stream that removes the
+        TOOL_PREFIX from tool names in the response body.
+        """
+        original_stream = response.stream
+        unprefix_fn = self._unprefix_tool_names_in_text
+
+        class UnprefixingStream(httpx.AsyncByteStream):
+            """Async byte stream that unprefixes tool names.
+
+            Inherits from httpx.AsyncByteStream to ensure proper stream interface.
+            """
+
+            def __init__(self, inner_stream: Any) -> None:
+                self._inner = inner_stream
+
+            async def __aiter__(self):
+                async for chunk in self._inner:
+                    if isinstance(chunk, bytes):
+                        text = chunk.decode("utf-8", errors="replace")
+                        text = unprefix_fn(text)
+                        yield text.encode("utf-8")
+                    else:
+                        yield chunk
+
+            async def aclose(self) -> None:
+                if hasattr(self._inner, "aclose"):
+                    try:
+                        result = self._inner.aclose()
+                        # Handle both sync and async aclose
+                        if hasattr(result, "__await__"):
+                            await result
+                    except Exception:
+                        pass  # Ignore close errors
+                elif hasattr(self._inner, "close"):
+                    try:
+                        self._inner.close()
+                    except Exception:
+                        pass
+
+        # Create a new response with the transformed stream
+        # Must include request for raise_for_status() to work
+        new_response = httpx.Response(
+            status_code=response.status_code,
+            headers=response.headers,
+            stream=UnprefixingStream(original_stream),
+            extensions=response.extensions,
+            request=request,
+        )
+        return new_response
+
     @staticmethod
     def _extract_body_bytes(request: httpx.Request) -> bytes | None:
         # Try public content first

code_puppy/command_line/core_commands.py

@@ -772,6 +772,91 @@ def handle_mcp_command(command: str) -> bool:
     return handler.handle_mcp_command(command)
 
 
+@register_command(
+    name="api",
+    description="Manage the Code Puppy API server",
+    usage="/api [start|stop|status]",
+    category="core",
+    detailed_help="Start, stop, or check status of the local FastAPI server for GUI integration.",
+)
+def handle_api_command(command: str) -> bool:
+    """Handle the /api command."""
+    import os
+    import signal
+    import subprocess
+    import sys
+    from pathlib import Path
+
+    from code_puppy.config import STATE_DIR
+    from code_puppy.messaging import emit_error, emit_info, emit_success
+
+    parts = command.split()
+    subcommand = parts[1] if len(parts) > 1 else "status"
+
+    pid_file = Path(STATE_DIR) / "api_server.pid"
+
+    if subcommand == "start":
+        # Check if already running
+        if pid_file.exists():
+            try:
+                pid = int(pid_file.read_text().strip())
+                os.kill(pid, 0)  # Check if process exists
+                emit_info(f"API server already running (PID {pid})")
+                return True
+            except (OSError, ValueError):
+                pid_file.unlink(missing_ok=True)  # Stale PID file
+
+        # Start the server in background
+        emit_info("Starting API server on http://127.0.0.1:8765 ...")
+        proc = subprocess.Popen(
+            [sys.executable, "-m", "code_puppy.api.main"],
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
+            start_new_session=True,
+        )
+        pid_file.parent.mkdir(parents=True, exist_ok=True)
+        pid_file.write_text(str(proc.pid))
+        emit_success(f"API server started (PID {proc.pid})")
+        emit_info("Docs available at http://127.0.0.1:8765/docs")
+        return True
+
+    elif subcommand == "stop":
+        if not pid_file.exists():
+            emit_info("API server is not running")
+            return True
+
+        try:
+            pid = int(pid_file.read_text().strip())
+            os.kill(pid, signal.SIGTERM)
+            pid_file.unlink()
+            emit_success(f"API server stopped (PID {pid})")
+        except (OSError, ValueError) as e:
+            pid_file.unlink(missing_ok=True)
+            emit_error(f"Error stopping server: {e}")
+        return True
+
+    elif subcommand == "status":
+        if not pid_file.exists():
+            emit_info("API server is not running")
+            return True
+
+        try:
+            pid = int(pid_file.read_text().strip())
+            os.kill(pid, 0)  # Check if process exists
+            emit_success(f"API server is running (PID {pid})")
+            emit_info("URL: http://127.0.0.1:8765")
+            emit_info("Docs: http://127.0.0.1:8765/docs")
+        except (OSError, ValueError):
+            pid_file.unlink(missing_ok=True)
+            emit_info("API server is not running (stale PID file removed)")
+        return True
+
+    else:
+        emit_error(f"Unknown subcommand: {subcommand}")
+        emit_info("Usage: /api [start|stop|status]")
+        return True
+
+
 @register_command(
     name="generate-pr-description",
     description="Generate comprehensive PR description",

code_puppy/config.py

@@ -75,6 +75,19 @@ def get_use_dbos() -> bool:
     return str(cfg_val).strip().lower() in {"1", "true", "yes", "on"}
 
 
+def get_subagent_verbose() -> bool:
+    """Return True if sub-agent verbose output is enabled (default False).
+
+    When False (default), sub-agents produce quiet, sparse output suitable
+    for parallel execution. When True, sub-agents produce full verbose output
+    like the main agent (useful for debugging).
+    """
+    cfg_val = get_value("subagent_verbose")
+    if cfg_val is None:
+        return False
+    return str(cfg_val).strip().lower() in {"1", "true", "yes", "on"}
+
+
 DEFAULT_SECTION = "puppy"
 REQUIRED_KEYS = ["puppy_name", "owner_name"]
 
@@ -85,7 +98,6 @@ _CURRENT_AUTOSAVE_ID: Optional[str] = None
 _model_validation_cache = {}
 _default_model_cache = None
 _default_vision_model_cache = None
-_default_vqa_model_cache = None
 
 
 def ensure_config_exists():
@@ -208,6 +220,9 @@ def get_config_keys():
         "diff_context_lines",
         "default_agent",
         "temperature",
+        "frontend_emitter_enabled",
+        "frontend_emitter_max_recent_events",
+        "frontend_emitter_queue_size",
     ]
     # Add DBOS control key
     default_keys.append("enable_dbos")
@@ -237,6 +252,22 @@ def set_config_value(key: str, value: str):
         config.write(f)
 
 
+# Alias for API compatibility
+def set_value(key: str, value: str) -> None:
+    """Set a config value. Alias for set_config_value."""
+    set_config_value(key, value)
+
+
+def reset_value(key: str) -> None:
+    """Remove a key from the config file, resetting it to default."""
+    config = configparser.ConfigParser()
+    config.read(CONFIG_FILE)
+    if DEFAULT_SECTION in config and key in config[DEFAULT_SECTION]:
+        del config[DEFAULT_SECTION][key]
+        with open(CONFIG_FILE, "w") as f:
+            config.write(f)
+
+
 # --- MODEL STICKY EXTENSION STARTS HERE ---
 def load_mcp_server_configs():
     """
@@ -326,47 +357,6 @@ def _default_vision_model_from_models_json() -> str:
         return "gpt-4.1"
 
 
-def _default_vqa_model_from_models_json() -> str:
-    """Select a default VQA-capable model, preferring vision-ready options."""
-    global _default_vqa_model_cache
-
-    if _default_vqa_model_cache is not None:
-        return _default_vqa_model_cache
-
-    try:
-        from code_puppy.model_factory import ModelFactory
-
-        models_config = ModelFactory.load_config()
-        if models_config:
-            # Allow explicit VQA hints if present
-            for name, config in models_config.items():
-                if config.get("supports_vqa"):
-                    _default_vqa_model_cache = name
-                    return name
-
-            # Reuse multimodal heuristics before falling back to generic default
-            preferred_candidates = (
-                "gpt-4.1",
-                "gpt-4.1-mini",
-                "claude-4-0-sonnet",
-                "gemini-2.5-flash-preview-05-20",
-                "gpt-4.1-nano",
-            )
-            for candidate in preferred_candidates:
-                if candidate in models_config:
-                    _default_vqa_model_cache = candidate
-                    return candidate
-
-            _default_vqa_model_cache = _default_model_from_models_json()
-            return _default_vqa_model_cache
-
-        _default_vqa_model_cache = "gpt-4.1"
-        return "gpt-4.1"
-    except Exception:
-        _default_vqa_model_cache = "gpt-4.1"
-        return "gpt-4.1"
-
-
 def _validate_model_exists(model_name: str) -> bool:
     """Check if a model exists in models.json with caching to avoid redundant calls."""
     global _model_validation_cache
@@ -392,15 +382,10 @@ def _validate_model_exists(model_name: str) -> bool:
 
 def clear_model_cache():
     """Clear the model validation cache. Call this when models.json changes."""
-    global \
-        _model_validation_cache, \
-        _default_model_cache, \
-        _default_vision_model_cache, \
-        _default_vqa_model_cache
+    global _model_validation_cache, _default_model_cache, _default_vision_model_cache
     _model_validation_cache.clear()
     _default_model_cache = None
     _default_vision_model_cache = None
-    _default_vqa_model_cache = None
 
 
 def model_supports_setting(model_name: str, setting: str) -> bool:
@@ -471,20 +456,6 @@ def set_model_name(model: str):
     clear_model_cache()
 
 
-def get_vqa_model_name() -> str:
-    """Return the configured VQA model, falling back to an inferred default."""
-    stored_model = get_value("vqa_model_name")
-    if stored_model and _validate_model_exists(stored_model):
-        return stored_model
-    return _default_vqa_model_from_models_json()
-
-
-def set_vqa_model_name(model: str):
-    """Persist the configured VQA model name and refresh caches."""
-    set_config_value("vqa_model_name", model or "")
-    clear_model_cache()
-
-
 def get_puppy_token():
     """Returns the puppy_token from config, or None if not set."""
     return get_value("puppy_token")
@@ -1291,6 +1262,8 @@ DEFAULT_BANNER_COLORS = {
     "invoke_agent": "deep_pink4",  # Ruby - agent invocation
     "subagent_response": "sea_green3",  # Emerald - sub-agent success
     "list_agents": "dark_slate_gray3",  # Slate - neutral listing
+    # Browser/Terminal tools - same color as edit_file (gold)
+    "terminal_tool": "dark_goldenrod",  # Gold - browser terminal operations
 }
 
 
@@ -1584,3 +1557,34 @@ def set_default_agent(agent_name: str) -> None:
         agent_name: The name of the agent to set as default.
     """
     set_config_value("default_agent", agent_name)
+
+
+# --- FRONTEND EMITTER CONFIGURATION ---
+def get_frontend_emitter_enabled() -> bool:
+    """Check if frontend emitter is enabled."""
+    val = get_value("frontend_emitter_enabled")
+    if val is None:
+        return True  # Enabled by default
+    return str(val).lower() in ("1", "true", "yes", "on")
+
+
+def get_frontend_emitter_max_recent_events() -> int:
+    """Get max number of recent events to buffer."""
+    val = get_value("frontend_emitter_max_recent_events")
+    if val is None:
+        return 100
+    try:
+        return int(val)
+    except ValueError:
+        return 100
+
+
+def get_frontend_emitter_queue_size() -> int:
+    """Get max subscriber queue size."""
+    val = get_value("frontend_emitter_queue_size")
+    if val is None:
+        return 100
+    try:
+        return int(val)
+    except ValueError:
+        return 100

code_puppy/messaging/__init__.py

@@ -113,6 +113,7 @@ from .messages import (  # Enums, Base, Text, File ops, Diff, Shell, Agent, etc.
     StatusPanelMessage,
     SubAgentInvocationMessage,
     SubAgentResponseMessage,
+    SubAgentStatusMessage,
     TextMessage,
     UserInputRequest,
     VersionCheckMessage,
@@ -120,6 +121,14 @@ from .messages import (  # Enums, Base, Text, File ops, Diff, Shell, Agent, etc.
 from .queue_console import QueueConsole, get_queue_console
 from .renderers import InteractiveRenderer, SynchronousInteractiveRenderer
 
+# Sub-agent console manager
+from .subagent_console import (
+    AgentState,
+    SubAgentConsoleManager,
+    get_subagent_console_manager,
+    STATUS_STYLES as SUBAGENT_STATUS_STYLES,
+)
+
 # Renderer
 from .rich_renderer import (
     DEFAULT_STYLES,
@@ -193,6 +202,7 @@ __all__ = [
     "AgentResponseMessage",
     "SubAgentInvocationMessage",
     "SubAgentResponseMessage",
+    "SubAgentStatusMessage",
     "UserInputRequest",
     "ConfirmationRequest",
     "SelectionRequest",
@@ -229,4 +239,9 @@ __all__ = [
     "DIFF_STYLES",
     # Markdown patches
     "patch_markdown_headings",
+    # Sub-agent console manager
+    "AgentState",
+    "SubAgentConsoleManager",
+    "get_subagent_console_manager",
+    "SUBAGENT_STATUS_STYLES",
 ]