code-puppy 0.0.335__py3-none-any.whl → 0.0.337__py3-none-any.whl

code_puppy/http_utils.py

@@ -5,10 +5,10 @@ This module provides functions for creating properly configured HTTP clients.
 """
 
 import asyncio
-import logging
 import os
 import socket
 import time
+from dataclasses import dataclass
 from typing import Any, Dict, Optional, Union
 
 import httpx
@@ -16,7 +16,69 @@ import requests
 
 from code_puppy.config import get_http2
 
-logger = logging.getLogger(__name__)
+
+@dataclass
+class ProxyConfig:
+    """Configuration for proxy and SSL settings."""
+
+    verify: Union[bool, str, None]
+    trust_env: bool
+    proxy_url: str | None
+    disable_retry: bool
+    http2_enabled: bool
+
+
+def _resolve_proxy_config(verify: Union[bool, str, None] = None) -> ProxyConfig:
+    """Resolve proxy, SSL, and retry settings from environment.
+
+    This centralizes the logic for detecting proxies, determining SSL verification,
+    and checking if retry transport should be disabled.
+    """
+    if verify is None:
+        verify = get_cert_bundle_path()
+
+    http2_enabled = get_http2()
+
+    disable_retry = os.environ.get(
+        "CODE_PUPPY_DISABLE_RETRY_TRANSPORT", ""
+    ).lower() in ("1", "true", "yes")
+
+    has_proxy = bool(
+        os.environ.get("HTTP_PROXY")
+        or os.environ.get("HTTPS_PROXY")
+        or os.environ.get("http_proxy")
+        or os.environ.get("https_proxy")
+    )
+
+    # Determine trust_env and verify based on proxy/retry settings
+    if disable_retry:
+        # Test mode: disable SSL verification for proxy testing
+        verify = False
+        trust_env = True
+    elif has_proxy:
+        # Production proxy: keep SSL verification enabled
+        trust_env = True
+    else:
+        trust_env = False
+
+    # Extract proxy URL
+    proxy_url = None
+    if has_proxy:
+        proxy_url = (
+            os.environ.get("HTTPS_PROXY")
+            or os.environ.get("https_proxy")
+            or os.environ.get("HTTP_PROXY")
+            or os.environ.get("http_proxy")
+        )
+
+    return ProxyConfig(
+        verify=verify,
+        trust_env=trust_env,
+        proxy_url=proxy_url,
+        disable_retry=disable_retry,
+        http2_enabled=http2_enabled,
+    )
+
 
 try:
     from .reopenable_async_client import ReopenableAsyncClient
@@ -58,14 +120,7 @@ class RetryingAsyncClient(httpx.AsyncClient):
 
         for attempt in range(self.max_retries + 1):
             try:
-                # Clone request for retry (streams might be consumed)
-                # But only if it's not the first attempt
-                req_to_send = request
-                if attempt > 0:
-                    # httpx requests are reusable, but we need to be careful with streams
-                    pass
-
-                response = await super().send(req_to_send, **kwargs)
+                response = await super().send(request, **kwargs)
                 last_response = response
 
                 # Check for retryable status
@@ -128,7 +183,7 @@ class RetryingAsyncClient(httpx.AsyncClient):
         return last_response
 
 
-def get_cert_bundle_path() -> str:
+def get_cert_bundle_path() -> str | None:
     # First check if SSL_CERT_FILE environment variable is set
     ssl_cert_file = os.environ.get("SSL_CERT_FILE")
     if ssl_cert_file and os.path.exists(ssl_cert_file):
@@ -164,66 +219,26 @@ def create_async_client(
     headers: Optional[Dict[str, str]] = None,
     retry_status_codes: tuple = (429, 502, 503, 504),
 ) -> httpx.AsyncClient:
-    if verify is None:
-        verify = get_cert_bundle_path()
-
-    # Check if HTTP/2 is enabled in config
-    http2_enabled = get_http2()
-
-    # Check if custom retry transport should be disabled (e.g., for integration tests with proxies)
-    disable_retry_transport = os.environ.get(
-        "CODE_PUPPY_DISABLE_RETRY_TRANSPORT", ""
-    ).lower() in ("1", "true", "yes")
-
-    # Check if proxy environment variables are set
-    has_proxy = bool(
-        os.environ.get("HTTP_PROXY")
-        or os.environ.get("HTTPS_PROXY")
-        or os.environ.get("http_proxy")
-        or os.environ.get("https_proxy")
-    )
-
-    # When retry transport is disabled (test mode), disable SSL verification
-    # for proxy testing. For production proxies, SSL should still be verified!
-    if disable_retry_transport:
-        verify = False
-        trust_env = True
-    elif has_proxy:
-        # Production proxy detected - keep SSL verification enabled for security
-        trust_env = True
-    else:
-        trust_env = False
-
-    # Extract proxy URL if needed
-    proxy_url = None
-    if has_proxy:
-        proxy_url = (
-            os.environ.get("HTTPS_PROXY")
-            or os.environ.get("https_proxy")
-            or os.environ.get("HTTP_PROXY")
-            or os.environ.get("http_proxy")
-        )
+    config = _resolve_proxy_config(verify)
 
-    # Use RetryingAsyncClient if retries are enabled
-    if not disable_retry_transport:
+    if not config.disable_retry:
         return RetryingAsyncClient(
             retry_status_codes=retry_status_codes,
-            proxy=proxy_url,
-            verify=verify,
+            proxy=config.proxy_url,
+            verify=config.verify,
             headers=headers or {},
             timeout=timeout,
-            http2=http2_enabled,
-            trust_env=trust_env,
+            http2=config.http2_enabled,
+            trust_env=config.trust_env,
         )
     else:
-        # Regular client for testing
         return httpx.AsyncClient(
-            proxy=proxy_url,
-            verify=verify,
+            proxy=config.proxy_url,
+            verify=config.verify,
             headers=headers or {},
             timeout=timeout,
-            http2=http2_enabled,
-            trust_env=trust_env,
+            http2=config.http2_enabled,
+            trust_env=config.trust_env,
         )
 
 
@@ -273,85 +288,33 @@ def create_reopenable_async_client(
     headers: Optional[Dict[str, str]] = None,
     retry_status_codes: tuple = (429, 502, 503, 504),
 ) -> Union[ReopenableAsyncClient, httpx.AsyncClient]:
-    if verify is None:
-        verify = get_cert_bundle_path()
-
-    # Check if HTTP/2 is enabled in config
-    http2_enabled = get_http2()
-
-    # Check if custom retry transport should be disabled (e.g., for integration tests with proxies)
-    disable_retry_transport = os.environ.get(
-        "CODE_PUPPY_DISABLE_RETRY_TRANSPORT", ""
-    ).lower() in ("1", "true", "yes")
-
-    # Check if proxy environment variables are set
-    has_proxy = bool(
-        os.environ.get("HTTP_PROXY")
-        or os.environ.get("HTTPS_PROXY")
-        or os.environ.get("http_proxy")
-        or os.environ.get("https_proxy")
-    )
-
-    # When retry transport is disabled (test mode), disable SSL verification
-    if disable_retry_transport:
-        verify = False
-        trust_env = True
-    elif has_proxy:
-        trust_env = True
-    else:
-        trust_env = False
+    config = _resolve_proxy_config(verify)
 
-    # Extract proxy URL if needed
-    proxy_url = None
-    if has_proxy:
-        proxy_url = (
-            os.environ.get("HTTPS_PROXY")
-            or os.environ.get("https_proxy")
-            or os.environ.get("HTTP_PROXY")
-            or os.environ.get("http_proxy")
-        )
+    base_kwargs = {
+        "proxy": config.proxy_url,
+        "verify": config.verify,
+        "headers": headers or {},
+        "timeout": timeout,
+        "http2": config.http2_enabled,
+        "trust_env": config.trust_env,
+    }
 
     if ReopenableAsyncClient is not None:
-        # Use RetryingAsyncClient if retries are enabled
         client_class = (
-            RetryingAsyncClient if not disable_retry_transport else httpx.AsyncClient
+            RetryingAsyncClient if not config.disable_retry else httpx.AsyncClient
         )
-
-        # Pass retry config only if using RetryingAsyncClient
-        kwargs = {
-            "proxy": proxy_url,
-            "verify": verify,
-            "headers": headers or {},
-            "timeout": timeout,
-            "http2": http2_enabled,
-            "trust_env": trust_env,
-        }
-
-        if not disable_retry_transport:
+        kwargs = {**base_kwargs, "client_class": client_class}
+        if not config.disable_retry:
             kwargs["retry_status_codes"] = retry_status_codes
-
-        return ReopenableAsyncClient(client_class=client_class, **kwargs)
+        return ReopenableAsyncClient(**kwargs)
     else:
-        # Fallback to RetryingAsyncClient
-        if not disable_retry_transport:
+        # Fallback to RetryingAsyncClient or plain AsyncClient
+        if not config.disable_retry:
             return RetryingAsyncClient(
-                retry_status_codes=retry_status_codes,
-                proxy=proxy_url,
-                verify=verify,
-                headers=headers or {},
-                timeout=timeout,
-                http2=http2_enabled,
-                trust_env=trust_env,
+                retry_status_codes=retry_status_codes, **base_kwargs
             )
         else:
-            return httpx.AsyncClient(
-                proxy=proxy_url,
-                verify=verify,
-                headers=headers or {},
-                timeout=timeout,
-                http2=http2_enabled,
-                trust_env=trust_env,
-            )
+            return httpx.AsyncClient(**base_kwargs)
 
 
 def is_cert_bundle_available() -> bool:

code_puppy/messaging/messages.py

@@ -209,6 +209,9 @@ class ShellStartMessage(BaseMessage):
         default=None, description="Working directory for the command"
     )
     timeout: int = Field(default=60, description="Timeout in seconds")
+    background: bool = Field(
+        default=False, description="Whether command runs in background mode"
+    )
 
 
 class ShellLineMessage(BaseMessage):

code_puppy/messaging/rich_renderer.py

@@ -348,7 +348,17 @@ class RichConsoleRenderer:
     # =========================================================================
 
     def _render_file_listing(self, msg: FileListingMessage) -> None:
-        """Render a directory listing matching the old Rich-formatted output."""
+        """Render a compact directory listing with directory summaries.
+
+        Instead of listing every file, we group by directory and show:
+        - Directory name
+        - Number of files
+        - Total size
+        - Number of subdirectories
+        """
+        import os
+        from collections import defaultdict
+
         # Header on single line
         rec_flag = f"(recursive={msg.recursive})"
         banner = self._format_banner("directory_listing", "DIRECTORY LISTING")
@@ -357,32 +367,104 @@ class RichConsoleRenderer:
             f"📂 [bold cyan]{msg.directory}[/bold cyan] [dim]{rec_flag}[/dim]\n"
         )
 
-        # Directory header
-        dir_name = msg.directory.rstrip("/").split("/")[-1] or msg.directory
-        self._console.print(f"📁 [bold blue]{dir_name}[/bold blue]")
+        # Build a tree structure: {parent_path: {files: [], dirs: set(), size: int}}
+        # Each key is a directory path, value contains direct children stats
+        dir_stats: dict = defaultdict(
+            lambda: {"files": [], "subdirs": set(), "total_size": 0}
+        )
+
+        # Root directory is represented as ""
+        root_key = ""
 
-        # Build tree structure from flat list
         for entry in msg.files:
-            # Calculate indentation based on depth
-            prefix = ""
-            for d in range(entry.depth + 1):
-                if d == entry.depth:
-                    prefix += "└── "
-                else:
-                    prefix += "    "
+            path = entry.path
+            parent = os.path.dirname(path) if os.path.dirname(path) else root_key
 
             if entry.type == "dir":
-                self._console.print(f"{prefix}📁 [bold blue]{entry.path}/[/bold blue]")
+                # Register this dir as a subdir of its parent
+                dir_stats[parent]["subdirs"].add(path)
+                # Ensure the dir itself exists in stats (even if empty)
+                _ = dir_stats[path]
             else:
-                icon = self._get_file_icon(entry.path)
-                if entry.size > 0:
-                    size_str = f" [dim]({self._format_size(entry.size)})[/dim]"
-                else:
-                    size_str = ""
+                # It's a file - add to parent's stats
+                dir_stats[parent]["files"].append(entry)
+                dir_stats[parent]["total_size"] += entry.size
+
+        def render_dir_tree(dir_path: str, depth: int = 0) -> None:
+            """Recursively render directory with compact summary."""
+            stats = dir_stats.get(
+                dir_path, {"files": [], "subdirs": set(), "total_size": 0}
+            )
+            files = stats["files"]
+            subdirs = sorted(stats["subdirs"])
+
+            # Calculate total size including subdirectories (recursive)
+            def get_recursive_size(d: str) -> int:
+                s = dir_stats.get(d, {"files": [], "subdirs": set(), "total_size": 0})
+                size = s["total_size"]
+                for sub in s["subdirs"]:
+                    size += get_recursive_size(sub)
+                return size
+
+            def get_recursive_file_count(d: str) -> int:
+                s = dir_stats.get(d, {"files": [], "subdirs": set(), "total_size": 0})
+                count = len(s["files"])
+                for sub in s["subdirs"]:
+                    count += get_recursive_file_count(sub)
+                return count
+
+            indent = "    " * depth
+
+            # For root level, just show contents
+            if dir_path == root_key:
+                # Show files at root level (depth 0)
+                for f in sorted(files, key=lambda x: x.path):
+                    icon = self._get_file_icon(f.path)
+                    name = os.path.basename(f.path)
+                    size_str = (
+                        f" [dim]({self._format_size(f.size)})[/dim]"
+                        if f.size > 0
+                        else ""
+                    )
+                    self._console.print(
+                        f"{indent}{icon} [green]{name}[/green]{size_str}"
+                    )
+
+                # Show subdirs at root level
+                for subdir in subdirs:
+                    render_dir_tree(subdir, depth)
+            else:
+                # Show directory with summary
+                dir_name = os.path.basename(dir_path)
+                rec_size = get_recursive_size(dir_path)
+                rec_file_count = get_recursive_file_count(dir_path)
+                subdir_count = len(subdirs)
+
+                # Build summary parts
+                parts = []
+                if rec_file_count > 0:
+                    parts.append(
+                        f"{rec_file_count} file{'s' if rec_file_count != 1 else ''}"
+                    )
+                if subdir_count > 0:
+                    parts.append(
+                        f"{subdir_count} subdir{'s' if subdir_count != 1 else ''}"
+                    )
+                if rec_size > 0:
+                    parts.append(self._format_size(rec_size))
+
+                summary = f" [dim]({', '.join(parts)})[/dim]" if parts else ""
                 self._console.print(
-                    f"{prefix}{icon} [green]{entry.path}[/green]{size_str}"
+                    f"{indent}📁 [bold blue]{dir_name}/[/bold blue]{summary}"
                 )
 
+                # Recursively show subdirectories
+                for subdir in subdirs:
+                    render_dir_tree(subdir, depth + 1)
+
+        # Render the tree starting from root
+        render_dir_tree(root_key, 0)
+
         # Summary
         self._console.print("\n[bold cyan]Summary:[/bold cyan]")
         self._console.print(
@@ -538,15 +620,25 @@ class RichConsoleRenderer:
         safe_command = escape_rich_markup(msg.command)
         # Header showing command is starting
         banner = self._format_banner("shell_command", "SHELL COMMAND")
-        self._console.print(f"\n{banner} 🚀 [dim]$ {safe_command}[/dim]")
+
+        # Add background indicator if running in background mode
+        if msg.background:
+            self._console.print(
+                f"\n{banner} 🚀 [dim]$ {safe_command}[/dim]  [bold magenta][BACKGROUND 🌙][/bold magenta]"
+            )
+        else:
+            self._console.print(f"\n{banner} 🚀 [dim]$ {safe_command}[/dim]")
 
         # Show working directory if specified
         if msg.cwd:
             safe_cwd = escape_rich_markup(msg.cwd)
             self._console.print(f"[dim]📂 Working directory: {safe_cwd}[/dim]")
 
-        # Show timeout
-        self._console.print(f"[dim]⏱ Timeout: {msg.timeout}s[/dim]")
+        # Show timeout or background status
+        if msg.background:
+            self._console.print("[dim]⏱ Runs detached (no timeout)[/dim]")
+        else:
+            self._console.print(f"[dim]⏱ Timeout: {msg.timeout}s[/dim]")
 
     def _render_shell_line(self, msg: ShellLineMessage) -> None:
         """Render shell output line preserving ANSI codes."""

code_puppy/plugins/antigravity_oauth/antigravity_model.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import base64
 import json
 import logging
 from collections.abc import AsyncIterator
@@ -54,7 +55,15 @@ class AntigravityModel(GoogleModel):
         messages: list[ModelMessage],
         model_request_parameters: ModelRequestParameters,
     ) -> tuple[ContentDict | None, list[dict]]:
-        """Map messages to Google GenAI format, preserving thinking signatures."""
+        """Map messages to Google GenAI format, preserving thinking signatures.
+
+        IMPORTANT: For Gemini with parallel function calls, the API expects:
+        - Model message: [FC1 + signature, FC2, ...] (all function calls together)
+        - User message: [FR1, FR2, ...] (all function responses together)
+
+        If messages are interleaved (FC1, FR1, FC2, FR2), the API returns 400.
+        This method merges consecutive same-role messages to fix this.
+        """
         contents: list[dict] = []
         system_parts: list[PartDict] = []
 
@@ -67,7 +76,16 @@ class AntigravityModel(GoogleModel):
                         system_parts.append({"text": part.content})
                     elif isinstance(part, UserPromptPart):
                         # Use parent's _map_user_prompt
-                        message_parts.extend(await self._map_user_prompt(part))
+                        mapped_parts = await self._map_user_prompt(part)
+                        # Sanitize bytes to base64 for JSON serialization
+                        for mp in mapped_parts:
+                            if "inline_data" in mp and "data" in mp["inline_data"]:
+                                data = mp["inline_data"]["data"]
+                                if isinstance(data, bytes):
+                                    mp["inline_data"]["data"] = base64.b64encode(
+                                        data
+                                    ).decode("utf-8")
+                        message_parts.extend(mapped_parts)
                     elif isinstance(part, ToolReturnPart):
                         message_parts.append(
                             {
@@ -95,7 +113,12 @@ class AntigravityModel(GoogleModel):
                         assert_never(part)
 
                 if message_parts:
-                    contents.append({"role": "user", "parts": message_parts})
+                    # Merge with previous user message if exists (for parallel function responses)
+                    if contents and contents[-1].get("role") == "user":
+                        contents[-1]["parts"].extend(message_parts)
+                    else:
+                        contents.append({"role": "user", "parts": message_parts})
+
             elif isinstance(m, ModelResponse):
                 # USE CUSTOM HELPER HERE
                 # Pass model name so we can handle Claude vs Gemini signature placement
@@ -103,7 +126,11 @@ class AntigravityModel(GoogleModel):
                     m, self.system, self._model_name
                 )
                 if maybe_content:
-                    contents.append(maybe_content)
+                    # Merge with previous model message if exists (for parallel function calls)
+                    if contents and contents[-1].get("role") == "model":
+                        contents[-1]["parts"].extend(maybe_content["parts"])
+                    else:
+                        contents.append(maybe_content)
             else:
                 assert_never(m)
 
@@ -435,6 +462,13 @@ class AntigravityStreamingResponse(StreamedResponse):
         return self._timestamp_val
 
 
+# Bypass signature for when no real thought signature is available.
+# Gemini API requires EVERY function call to have a thoughtSignature field.
+# When there's no thinking block or no signature was captured, we use this bypass.
+# This specific key is the official bypass token for Gemini 3 Pro.
+BYPASS_THOUGHT_SIGNATURE = "context_engineering_is_the_way_to_go"
+
+
 def _antigravity_content_model_response(
     m: ModelResponse, provider_name: str, model_name: str = ""
 ) -> ContentDict | None:
@@ -443,6 +477,10 @@ def _antigravity_content_model_response(
     Handles different signature protocols:
     - Claude models: signature goes ON the thinking block itself
     - Gemini models: signature goes on the NEXT part (function_call or text) after thinking
+
+    IMPORTANT: For Gemini, EVERY function call MUST have a thoughtSignature field.
+    If no real signature is available (no preceding ThinkingPart, or ThinkingPart
+    had no signature), we use BYPASS_THOUGHT_SIGNATURE as a fallback.
     """
     parts: list[PartDict] = []
 
@@ -451,6 +489,7 @@ def _antigravity_content_model_response(
     is_gemini = "gemini" in model_name.lower()
 
     # For Gemini: save signature from ThinkingPart to attach to next part
+    # Initialize to None - we'll use BYPASS_THOUGHT_SIGNATURE if still None when needed
     pending_signature: str | None = None
 
     for item in m.parts:
@@ -462,16 +501,24 @@ def _antigravity_content_model_response(
             )
             part["function_call"] = function_call
 
-            # For Gemini: attach pending signature to function call
-            if is_gemini and pending_signature:
-                part["thoughtSignature"] = pending_signature
-                pending_signature = None
+            # For Gemini: ALWAYS attach a thoughtSignature to function calls.
+            # Use the real signature if available, otherwise use bypass.
+            # NOTE: Do NOT clear pending_signature here! Multiple tool calls
+            # in a row (e.g., parallel function calls) all need the same
+            # signature from the preceding ThinkingPart.
+            if is_gemini:
+                part["thoughtSignature"] = (
+                    pending_signature
+                    if pending_signature is not None
+                    else BYPASS_THOUGHT_SIGNATURE
+                )
 
         elif isinstance(item, TextPart):
             part["text"] = item.content
 
-            # For Gemini: attach pending signature to text part
-            if is_gemini and pending_signature:
+            # For Gemini: attach pending signature to text part if available
+            # Clear signature after text since text typically ends a response
+            if is_gemini and pending_signature is not None:
                 part["thoughtSignature"] = pending_signature
                 pending_signature = None
 
@@ -490,6 +537,10 @@ def _antigravity_content_model_response(
                     else:
                         # Default: try both (put on thinking block)
                         part["thoughtSignature"] = item.signature
+                elif is_gemini:
+                    # ThinkingPart exists but has no signature - use bypass
+                    # This ensures subsequent tool calls still get a signature
+                    pending_signature = BYPASS_THOUGHT_SIGNATURE
 
         elif isinstance(item, BuiltinToolCallPart):
             # Skip code execution for now
@@ -501,8 +552,13 @@ def _antigravity_content_model_response(
 
         elif isinstance(item, FilePart):
             content = item.content
+            # Ensure data is base64 string, not bytes
+            data_val = content.data
+            if isinstance(data_val, bytes):
+                data_val = base64.b64encode(data_val).decode("utf-8")
+
             inline_data_dict: BlobDict = {
-                "data": content.data,
+                "data": data_val,
                 "mime_type": content.media_type,
             }
             part["inline_data"] = inline_data_dict

code_puppy/plugins/antigravity_oauth/transport.py

@@ -422,6 +422,8 @@ class AntigravityClient(httpx.AsyncClient):
 
     async def send(self, request: httpx.Request, **kwargs: Any) -> httpx.Response:
         """Override send to intercept at the lowest level with endpoint fallback."""
+        import asyncio
+
         # Transform POST requests to Antigravity format
         if request.method == "POST" and request.content:
             new_content, new_path, new_query, is_claude_thinking = self._wrap_request(
@@ -467,25 +469,21 @@ class AntigravityClient(httpx.AsyncClient):
                     else:
                         new_headers["anthropic-beta"] = interleaved_header
 
-                # Try each endpoint in fallback order
+                # Try each endpoint with rate limit retry logic
                 last_response = None
-
-                # Check for rate limit
-                import asyncio
-
-                # Try endpoints logic...
-                max_retries = 3
-
-                for attempt in range(max_retries):
-                    for endpoint in ANTIGRAVITY_ENDPOINT_FALLBACKS:
-                        # Build URL with current endpoint
-                        new_url = httpx.URL(
-                            scheme="https",
-                            host=endpoint.replace("https://", ""),
-                            path=new_path,
-                            query=new_query.encode() if new_query else b"",
-                        )
-
+                max_rate_limit_retries = 5  # Max retries for 429s per endpoint
+
+                for endpoint in ANTIGRAVITY_ENDPOINT_FALLBACKS:
+                    # Build URL with current endpoint
+                    new_url = httpx.URL(
+                        scheme="https",
+                        host=endpoint.replace("https://", ""),
+                        path=new_path,
+                        query=new_query.encode() if new_query else b"",
+                    )
+
+                    # Retry loop for rate limits on this endpoint
+                    for rate_limit_attempt in range(max_rate_limit_retries):
                         req = httpx.Request(
                             method=request.method,
                             url=new_url,
@@ -498,45 +496,35 @@ class AntigravityClient(httpx.AsyncClient):
 
                         # Handle rate limit (429)
                         if response.status_code == 429:
-                            try:
-                                # Need to read response to get reset time
-                                await response.aread()
-                                error_data = json.loads(response.content)
-
-                                # Extract reset delay from metadata
-                                reset_delay = 2.0  # Default fallback
-                                if isinstance(error_data, dict):
-                                    error_info = error_data.get("error", {})
-                                    if isinstance(error_info, dict):
-                                        details = error_info.get("details", [])
-                                        for detail in details:
-                                            metadata = detail.get("metadata", {})
-                                            if "quotaResetDelay" in metadata:
-                                                delay_str = metadata["quotaResetDelay"]
-                                                if delay_str.endswith("s"):
-                                                    reset_delay = float(delay_str[:-1])
-                                                break
-
-                                # Only retry if delay is reasonable (< 60s)
-                                if reset_delay < 60:
-                                    # Add small buffer
-                                    wait_time = reset_delay + 0.5
+                            wait_time = await self._extract_rate_limit_delay(response)
+
+                            if wait_time is not None and wait_time < 60:
+                                # Add small buffer to wait time
+                                wait_time = wait_time + 0.1
+                                try:
                                     from code_puppy.messaging import emit_warning
 
                                     emit_warning(
-                                        f"⏳ Rate limited on {endpoint}. Waiting {wait_time:.1f}s..."
+                                        f"⏳ Rate limited (attempt {rate_limit_attempt + 1}/{max_rate_limit_retries}). "
+                                        f"Waiting {wait_time:.2f}s..."
                                     )
-                                    await asyncio.sleep(wait_time)
-                                    continue  # Retry same endpoint
-                                else:
-                                    emit_warning(
-                                        f"❌ Rate limit reset too long ({reset_delay}s). Aborting."
+                                except ImportError:
+                                    logger.warning(
+                                        "Rate limited, waiting %.2fs...", wait_time
                                     )
-                                    return UnwrappedResponse(response)
-                            except Exception:
-                                pass  # Failed to parse error, treat as normal error
 
-                        # Retry on 403, 404, 5xx errors
+                                await asyncio.sleep(wait_time)
+                                continue  # Retry same endpoint
+                            else:
+                                # Wait time too long or couldn't parse, try next endpoint
+                                logger.debug(
+                                    "Rate limit wait too long (%.1fs) on %s, trying next endpoint...",
+                                    wait_time or 0,
+                                    endpoint,
+                                )
+                                break  # Break inner loop, try next endpoint
+
+                        # Retry on 403, 404, 5xx errors - try next endpoint
                         if (
                             response.status_code in (403, 404)
                             or response.status_code >= 500
@@ -546,9 +534,9 @@ class AntigravityClient(httpx.AsyncClient):
                                 endpoint,
                                 response.status_code,
                             )
-                            continue
+                            break  # Try next endpoint
 
-                        # Success or non-retriable error
+                        # Success or non-retriable error (4xx except 429)
                         # Wrap response to unwrap Antigravity format
                         if "alt=sse" in new_query:
                             return UnwrappedSSEResponse(response)
@@ -558,15 +546,96 @@ class AntigravityClient(httpx.AsyncClient):
                         await response.aread()
                         return UnwrappedResponse(response)
 
-                # All endpoints failed, return last response
-                if last_response and last_response.status_code == 429:
-                    await last_response.aread()
+                # All endpoints/retries exhausted, return last response
+                if last_response:
+                    # Ensure response is read for proper error handling
+                    if not last_response.is_stream_consumed:
+                        try:
+                            await last_response.aread()
+                        except Exception:
+                            pass
                     return UnwrappedResponse(last_response)
 
-                return last_response
-
         return await super().send(request, **kwargs)
 
+    async def _extract_rate_limit_delay(self, response: httpx.Response) -> float | None:
+        """Extract the retry delay from a 429 rate limit response.
+
+        Parses the Antigravity/Google API error format to find:
+        - retryDelay from RetryInfo (e.g., "0.088325827s")
+        - quotaResetDelay from ErrorInfo metadata (e.g., "88.325827ms")
+
+        Returns the delay in seconds, or None if parsing fails.
+        """
+        try:
+            # Read response body if not already read
+            if not response.is_stream_consumed:
+                await response.aread()
+
+            error_data = json.loads(response.content)
+
+            if not isinstance(error_data, dict):
+                return 2.0  # Default fallback
+
+            error_info = error_data.get("error", {})
+            if not isinstance(error_info, dict):
+                return 2.0
+
+            details = error_info.get("details", [])
+            if not isinstance(details, list):
+                return 2.0
+
+            # Look for RetryInfo first (most precise)
+            for detail in details:
+                if not isinstance(detail, dict):
+                    continue
+
+                detail_type = detail.get("@type", "")
+
+                # Check for RetryInfo (e.g., "0.088325827s")
+                if "RetryInfo" in detail_type:
+                    retry_delay = detail.get("retryDelay", "")
+                    parsed = self._parse_duration(retry_delay)
+                    if parsed is not None:
+                        return parsed
+
+                # Check for ErrorInfo with quotaResetDelay in metadata
+                if "ErrorInfo" in detail_type:
+                    metadata = detail.get("metadata", {})
+                    if isinstance(metadata, dict):
+                        quota_delay = metadata.get("quotaResetDelay", "")
+                        parsed = self._parse_duration(quota_delay)
+                        if parsed is not None:
+                            return parsed
+
+            return 2.0  # Default if no delay found
+
+        except (json.JSONDecodeError, Exception) as e:
+            logger.debug("Failed to parse rate limit response: %s", e)
+            return 2.0  # Default fallback
+
+    def _parse_duration(self, duration_str: str) -> float | None:
+        """Parse a duration string like '0.088s' or '88.325827ms' to seconds."""
+        if not duration_str or not isinstance(duration_str, str):
+            return None
+
+        duration_str = duration_str.strip()
+
+        try:
+            # Handle milliseconds (e.g., "88.325827ms")
+            if duration_str.endswith("ms"):
+                return float(duration_str[:-2]) / 1000.0
+
+            # Handle seconds (e.g., "0.088325827s")
+            if duration_str.endswith("s"):
+                return float(duration_str[:-1])
+
+            # Try parsing as raw number (assume seconds)
+            return float(duration_str)
+
+        except ValueError:
+            return None
+
 
 def create_antigravity_client(
     access_token: str,

code_puppy/tools/command_runner.py

@@ -880,6 +880,7 @@ async def run_shell_command(
                     command=command,
                     cwd=cwd,
                     timeout=0,  # No timeout for background processes
+                    background=True,
                 )
             )
 

{code_puppy-0.0.335.dist-info → code_puppy-0.0.337.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-puppy
-Version: 0.0.335
+Version: 0.0.337
 Summary: Code generation agent
 Project-URL: repository, https://github.com/mpfaffenberger/code_puppy
 Project-URL: HomePage, https://github.com/mpfaffenberger/code_puppy

{code_puppy-0.0.335.dist-info → code_puppy-0.0.337.dist-info}/RECORD

@@ -7,7 +7,7 @@ code_puppy/cli_runner.py,sha256=4iosJ_zXv9WcG4764yFW-VOLjE3B7Og9yclp6Q4kaSQ,3387
 code_puppy/config.py,sha256=RlnrLkyFXm7h2Htf8rQA7vqoAyzLPMrESle417uLmFw,52373
 code_puppy/error_logging.py,sha256=a80OILCUtJhexI6a9GM-r5LqIdjvSRzggfgPp2jv1X0,3297
 code_puppy/gemini_code_assist.py,sha256=KGS7sO5OLc83nDF3xxS-QiU6vxW9vcm6hmzilu79Ef8,13867
-code_puppy/http_utils.py,sha256=BV_bf0i5iPh7HC7T1vlsbz3hvlVwdIzwpS2pfBBKPRo,11823
+code_puppy/http_utils.py,sha256=H3N5Qz2B1CcsGUYOycGWAqoNMr2P1NCVluKX3aRwRqI,10358
 code_puppy/keymap.py,sha256=IvMkTlB_bIqOWpbTpmftkdyjhtD5todXuEIw1zCZ4u0,3584
 code_puppy/main.py,sha256=82r3vZy_XcyEsenLn82BnUusaoyL3Bpm_Th_jKgqecE,273
 code_puppy/model_factory.py,sha256=Djhp_ukLTMNi8UdZsodhIU6D2X4DmJXnjZy_bzfce3k,37517
@@ -112,10 +112,10 @@ code_puppy/messaging/bus.py,sha256=TbdltJ0D5tqnaE4irq1fcXllDYm-mQ_SiX1IFm-S4sw,2
 code_puppy/messaging/commands.py,sha256=77CtKVNaF5KS3Xyzd0ccDAisZWQxL3weVEt3J-SfYxo,5464
 code_puppy/messaging/markdown_patches.py,sha256=dMIJozzJChuHa8QNMSEz_kC-dyt7kZiDLZ7rjthbcmg,1626
 code_puppy/messaging/message_queue.py,sha256=e-viZxacBoNSxRJnCJ4hU4vzsSI3oX_rN58RwhJKFfU,11825
-code_puppy/messaging/messages.py,sha256=vNenUTgeZLk2clH4-BLcj44vpzDMmAsOrbwn_msaCVw,16351
+code_puppy/messaging/messages.py,sha256=F7RwMHeQrIk-8kuSSBU76wBq1NGuLb2H5cJrSMTC3XM,16464
 code_puppy/messaging/queue_console.py,sha256=T0U_V1tdN6hd9DLokp-HCk0mhu8Ivpfajha368CBZrU,9983
 code_puppy/messaging/renderers.py,sha256=GHVtMnxE1pJ-yrcRjacY81JcjlHRz3UVHzp-ohN-CGE,12058
-code_puppy/messaging/rich_renderer.py,sha256=Kln4L3Lt-KD0orCav9CBxK0Ggz1u8eV4HbohcUvyENk,34215
+code_puppy/messaging/rich_renderer.py,sha256=FiT1e5S8nNQte0E6CMFQ3KyTixadkgKSjp1hcZXtyOE,37892
 code_puppy/messaging/spinner/__init__.py,sha256=KpK5tJqq9YnN3wklqvdH0BQmuwYnT83Mp4tPfQa9RqI,1664
 code_puppy/messaging/spinner/console_spinner.py,sha256=YIReuWPD01YPy58FqWdMDWj2QhauTUxKo675Ub4-eDA,8451
 code_puppy/messaging/spinner/spinner_base.py,sha256=JiQDAhCfwrWUFunb8Xcj1caEl34JJY7Bcio7mDeckSc,2694
@@ -123,7 +123,7 @@ code_puppy/plugins/__init__.py,sha256=gWgrXWoFpl-3Mxz2DAvxKW6SkCWrOnw-hKsY9O7nHc
 code_puppy/plugins/oauth_puppy_html.py,sha256=Wpa-V_NlRiBAvo_OXHuR7wvOH_jSt8L9HSFGiab6xI0,13058
 code_puppy/plugins/antigravity_oauth/__init__.py,sha256=1miHihSqRNXO20Vh_Gn9M3Aa2szh0gtdSCaKKj9nq0Q,362
 code_puppy/plugins/antigravity_oauth/accounts.py,sha256=GQit2-K24bsopmTZyscFUq3M0cAEO5WutHWnipVdgz8,14304
-code_puppy/plugins/antigravity_oauth/antigravity_model.py,sha256=FxL2MPLLoA8MpxxAGh6GqUAqUZiUSNRt16jU4YpaKVk,22984
+code_puppy/plugins/antigravity_oauth/antigravity_model.py,sha256=g0_nXnMg288CvBE48CFgZ-iAqlYbtbi1dcB4Up6cYYc,26115
 code_puppy/plugins/antigravity_oauth/config.py,sha256=BoQgqf5I2XoHWnBBo9vhCIc_XwPj9Mbp0Z95ygWwt78,1362
 code_puppy/plugins/antigravity_oauth/constants.py,sha256=qsrA10JJvzNuY0OobvvwCQcoGpILBninllcUUMKkUrQ,4644
 code_puppy/plugins/antigravity_oauth/oauth.py,sha256=ZHXJtZP63l6brOpX1WdLfuUClIleA79-4y36YUJc6Wo,15137
@@ -131,7 +131,7 @@ code_puppy/plugins/antigravity_oauth/register_callbacks.py,sha256=uKIvfzH-dXj1g_
 code_puppy/plugins/antigravity_oauth/storage.py,sha256=LW1DkY6Z-GRbBDrIitT6glKemZptp3NzldIrLRqTAK0,8971
 code_puppy/plugins/antigravity_oauth/test_plugin.py,sha256=n0kjFG8Vt2n1j0GgTRSdSyhF0t9xxE8Ht60SH5CSwzw,11027
 code_puppy/plugins/antigravity_oauth/token.py,sha256=WbiFCkrZvChpGXvwIYsJMgqU9xdJ81KwR062lFlnL3U,5038
-code_puppy/plugins/antigravity_oauth/transport.py,sha256=FpP4NYCA3tdSISCPQTcmtFWGvLeLH-MdvpwjG8LY7wQ,25228
+code_puppy/plugins/antigravity_oauth/transport.py,sha256=yZztRm8NHWemAtv7aVKsHdCQtU9BJKAd9RcqF91ZQOw,27689
 code_puppy/plugins/antigravity_oauth/utils.py,sha256=mXHRv0l07r27VjtSsIy9rlpkUheP88RaM4x4M0O1mMY,5401
 code_puppy/plugins/chatgpt_oauth/__init__.py,sha256=Kjc6Hsz1sWvMD2OdAlWZvJRiKJSj4fx22boa-aVFKjA,189
 code_puppy/plugins/chatgpt_oauth/config.py,sha256=H_wAH9Duyn8WH2Kq8oe72uda-_4qu1uXLPun_SDdtsk,2023
@@ -159,7 +159,7 @@ code_puppy/plugins/shell_safety/register_callbacks.py,sha256=W3v664RR48Fdbbbltf_
 code_puppy/prompts/codex_system_prompt.md,sha256=hEFTCziroLqZmqNle5kG34A8kvTteOWezCiVrAEKhE0,24400
 code_puppy/tools/__init__.py,sha256=BVTZ85jLHgDANwOnUSOz3UDlp8VQDq4DoGF23BRlyWw,6032
 code_puppy/tools/agent_tools.py,sha256=snBI6FlFtR03CbYKXwu53R48c_fRSuDIwcNdVUruLcA,21020
-code_puppy/tools/command_runner.py,sha256=4RlgPWWE2Mk35t69QJWR4Md4etMGLR2rFSZAKrNohU4,44917
+code_puppy/tools/command_runner.py,sha256=3qXVnVTaBPia6y2D29As47_TRKgpyCj82yMFK-8UUYc,44954
 code_puppy/tools/common.py,sha256=IboS6sbwN4a3FzHdfsZJtEFiyDUCszevI6LpH14ydEk,40561
 code_puppy/tools/file_modifications.py,sha256=vz9n7R0AGDSdLUArZr_55yJLkyI30M8zreAppxIx02M,29380
 code_puppy/tools/file_operations.py,sha256=CqhpuBnOFOcQCIYXOujskxq2VMLWYJhibYrH0YcPSfA,35692
@@ -174,10 +174,10 @@ code_puppy/tools/browser/browser_scripts.py,sha256=sNb8eLEyzhasy5hV4B9OjM8yIVMLV
 code_puppy/tools/browser/browser_workflows.py,sha256=nitW42vCf0ieTX1gLabozTugNQ8phtoFzZbiAhw1V90,6491
 code_puppy/tools/browser/camoufox_manager.py,sha256=RZjGOEftE5sI_tsercUyXFSZI2wpStXf-q0PdYh2G3I,8680
 code_puppy/tools/browser/vqa_agent.py,sha256=DBn9HKloILqJSTSdNZzH_PYWT0B2h9VwmY6akFQI_uU,2913
-code_puppy-0.0.335.data/data/code_puppy/models.json,sha256=IPABdOrDw2OZJxa0XGBWSWmBRerV6_pIEmKVLRtUbAk,3105
-code_puppy-0.0.335.data/data/code_puppy/models_dev_api.json,sha256=wHjkj-IM_fx1oHki6-GqtOoCrRMR0ScK0f-Iz0UEcy8,548187
-code_puppy-0.0.335.dist-info/METADATA,sha256=ORG4lkEhizZykfuFWcudUzIkPCG0ogmStOS8q0NZ6IA,28854
-code_puppy-0.0.335.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-code_puppy-0.0.335.dist-info/entry_points.txt,sha256=Tp4eQC99WY3HOKd3sdvb22vZODRq0XkZVNpXOag_KdI,91
-code_puppy-0.0.335.dist-info/licenses/LICENSE,sha256=31u8x0SPgdOq3izJX41kgFazWsM43zPEF9eskzqbJMY,1075
-code_puppy-0.0.335.dist-info/RECORD,,
+code_puppy-0.0.337.data/data/code_puppy/models.json,sha256=IPABdOrDw2OZJxa0XGBWSWmBRerV6_pIEmKVLRtUbAk,3105
+code_puppy-0.0.337.data/data/code_puppy/models_dev_api.json,sha256=wHjkj-IM_fx1oHki6-GqtOoCrRMR0ScK0f-Iz0UEcy8,548187
+code_puppy-0.0.337.dist-info/METADATA,sha256=F1DCPuk3QS0E2t1o_HRId-DiNTtOs29kcmSueGwHUHk,28854
+code_puppy-0.0.337.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+code_puppy-0.0.337.dist-info/entry_points.txt,sha256=Tp4eQC99WY3HOKd3sdvb22vZODRq0XkZVNpXOag_KdI,91
+code_puppy-0.0.337.dist-info/licenses/LICENSE,sha256=31u8x0SPgdOq3izJX41kgFazWsM43zPEF9eskzqbJMY,1075
+code_puppy-0.0.337.dist-info/RECORD,,