code-puppy 0.0.371__tar.gz → 0.0.373__tar.gz

{code_puppy-0.0.371 → code_puppy-0.0.373}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-puppy
-Version: 0.0.371
+Version: 0.0.373
 Summary: Code generation agent
 Project-URL: repository, https://github.com/mpfaffenberger/code_puppy
 Project-URL: HomePage, https://github.com/mpfaffenberger/code_puppy

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/claude_cache_client.py

@@ -26,7 +26,7 @@ import httpx
 
 logger = logging.getLogger(__name__)
 
-# Refresh token if it's older than 1 hour (3600 seconds)
+# Refresh token if it's older than the configured max age (seconds)
 TOKEN_MAX_AGE_SECONDS = 3600
 
 # Tool name prefix for Claude Code OAuth compatibility
@@ -94,13 +94,13 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
                 return age
 
             # Fall back to calculating from 'exp' claim
-            # Assume tokens are typically valid for 1 hour
+            # Assume tokens are typically valid for TOKEN_MAX_AGE_SECONDS
             if "exp" in payload:
                 exp = float(payload["exp"])
                 # If exp is in the future, calculate how long until expiry
-                # and assume the token was issued 1 hour before expiry
+                # and assume the token was issued TOKEN_MAX_AGE_SECONDS before expiry
                 time_until_exp = exp - now
-                # If token has less than 1 hour left, it's "old"
+                # If token has less than TOKEN_MAX_AGE_SECONDS left, it's "old"
                 age = TOKEN_MAX_AGE_SECONDS - time_until_exp
                 return max(0, age)
 
@@ -119,13 +119,13 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
         return None
 
     def _should_refresh_token(self, request: httpx.Request) -> bool:
-        """Check if the token should be refreshed (within 1 hour of expiry).
+        """Check if the token should be refreshed (within the max-age window).
 
         Uses two strategies:
         1. Decode JWT to check token age (if possible)
         2. Fall back to stored expires_at from token file
 
-        Returns True if token expires within TOKEN_MAX_AGE_SECONDS (1 hour).
+        Returns True if token expires within TOKEN_MAX_AGE_SECONDS.
         """
         token = self._extract_bearer_token(request)
         if not token:
@@ -169,7 +169,7 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
             if not tokens:
                 return False
 
-            # is_token_expired already uses TOKEN_REFRESH_BUFFER_SECONDS (1 hour)
+            # is_token_expired already uses the configured refresh buffer window
             return is_token_expired(tokens)
         except Exception as exc:
             logger.debug("Error checking stored token expiry: %s", exc)
@@ -366,10 +366,10 @@ class ClaudeCacheAsyncClient(httpx.AsyncClient):
 
         # Handle auth errors with token refresh
         try:
-            if response.status_code in (400, 401) and not request.extensions.get(
+            if response.status_code in (400, 401, 403) and not request.extensions.get(
                 "claude_oauth_refresh_attempted"
             ):
-                is_auth_error = response.status_code == 401
+                is_auth_error = response.status_code in (401, 403)
 
                 if response.status_code == 400:
                     is_auth_error = self._is_cloudflare_html_error(response)

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/command_line/add_model_menu.py

@@ -626,12 +626,21 @@ class AddModelMenu:
         elif model_type == "openai" and "gpt-5" in model.model_id:
             # GPT-5 models have special settings
             if "codex" in model.model_id:
-                config["supported_settings"] = ["reasoning_effort"]
+                config["supported_settings"] = [
+                    "temperature",
+                    "top_p",
+                    "reasoning_effort",
+                ]
             else:
-                config["supported_settings"] = ["reasoning_effort", "verbosity"]
+                config["supported_settings"] = [
+                    "temperature",
+                    "top_p",
+                    "reasoning_effort",
+                    "verbosity",
+                ]
         else:
-            # Default settings for most models (no top_p)
-            config["supported_settings"] = ["temperature", "seed"]
+            # Default settings for most models
+            config["supported_settings"] = ["temperature", "seed", "top_p"]
 
         return config
 

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/command_line/model_settings_menu.py

@@ -39,10 +39,10 @@ SETTING_DEFINITIONS: Dict[str, Dict] = {
         "description": "Controls randomness. Lower = more deterministic, higher = more creative.",
         "type": "numeric",
         "min": 0.0,
-        "max": 1.0,  # Clamped to 0-1 per user request
-        "step": 0.1,
+        "max": 1.0,
+        "step": 0.05,
         "default": None,  # None means use model default
-        "format": "{:.1f}",
+        "format": "{:.2f}",
     },
     "seed": {
         "name": "Seed",
@@ -54,6 +54,16 @@ SETTING_DEFINITIONS: Dict[str, Dict] = {
         "default": None,
         "format": "{:.0f}",
     },
+    "top_p": {
+        "name": "Top-P (Nucleus Sampling)",
+        "description": "Controls token diversity. 0.0 = least random (only most likely tokens), 1.0 = most random (sample from all tokens).",
+        "type": "numeric",
+        "min": 0.0,
+        "max": 1.0,
+        "step": 0.05,
+        "default": None,
+        "format": "{:.2f}",
+    },
     "reasoning_effort": {
         "name": "Reasoning Effort",
         "description": "Controls how much effort GPT-5 models spend on reasoning. Higher = more thorough but slower.",
@@ -575,6 +585,8 @@ class ModelSettingsMenu:
             # Default to a sensible starting point for numeric
             if setting_key == "temperature":
                 self.edit_value = 0.7
+            elif setting_key == "top_p":
+                self.edit_value = 0.9  # Common default for top_p
             elif setting_key == "seed":
                 self.edit_value = 42
             elif setting_key == "budget_tokens":

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/models.json

@@ -7,7 +7,7 @@
       "api_key": "$SYN_API_KEY"
     },
     "context_length": 200000,
-    "supported_settings": ["temperature", "seed"]
+    "supported_settings": ["temperature", "seed", "top_p"]
   },
   "synthetic-MiniMax-M2.1": {
     "type": "custom_openai",
@@ -17,7 +17,7 @@
       "api_key": "$SYN_API_KEY"
     },
     "context_length": 195000,
-    "supported_settings": ["temperature", "seed"]
+    "supported_settings": ["temperature", "seed", "top_p"]
   },
   "synthetic-Kimi-K2-Thinking": {
     "type": "custom_openai",
@@ -27,32 +27,32 @@
       "api_key": "$SYN_API_KEY"
     },
     "context_length": 262144,
-    "supported_settings": ["temperature", "seed"]
+    "supported_settings": ["temperature", "seed", "top_p"]
   },
   "Gemini-3": {
     "type": "gemini",
     "name": "gemini-3-pro-preview",
     "context_length": 200000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   },
   "Gemini-3-Long-Context": {
     "type": "gemini",
     "name": "gemini-3-pro-preview",
     "context_length": 1000000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   },
   "gpt-5.1": {
     "type": "openai",
     "name": "gpt-5.1",
     "context_length": 272000,
-    "supported_settings": ["reasoning_effort", "verbosity"],
+    "supported_settings": ["temperature", "top_p", "reasoning_effort", "verbosity"],
     "supports_xhigh_reasoning": false
   },
   "gpt-5.1-codex-api": {
     "type": "openai",
     "name": "gpt-5.1-codex",
     "context_length": 272000,
-    "supported_settings": ["reasoning_effort", "verbosity"],
+    "supported_settings": ["temperature", "top_p", "reasoning_effort", "verbosity"],
     "supports_xhigh_reasoning": true
   },
   "Cerebras-GLM-4.7": {
@@ -63,7 +63,7 @@
       "api_key": "$CEREBRAS_API_KEY"
     },
     "context_length": 131072,
-    "supported_settings": ["temperature", "seed"]
+    "supported_settings": ["temperature", "seed", "top_p"]
   },
   "claude-4-5-haiku": {
     "type": "anthropic",
@@ -87,24 +87,24 @@
     "type": "zai_coding",
     "name": "glm-4.6",
     "context_length": 200000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   },
   "zai-glm-4.6-api": {
     "type": "zai_api",
     "name": "glm-4.6",
     "context_length": 200000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   },
   "zai-glm-4.7-coding": {
     "type": "zai_coding",
     "name": "glm-4.7",
     "context_length": 200000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   },
   "zai-glm-4.7-api": {
     "type": "zai_api",
     "name": "glm-4.7",
     "context_length": 200000,
-    "supported_settings": ["temperature"]
+    "supported_settings": ["temperature", "top_p"]
   }
 }

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/plugins/claude_code_oauth/README.md

@@ -52,7 +52,7 @@ The plugin ships with sensible defaults in `config.py`:
 ```python
 CLAUDE_CODE_OAUTH_CONFIG = {
     "auth_url": "https://claude.ai/oauth/authorize",
-    "token_url": "https://claude.ai/api/oauth/token",
+    "token_url": "https://console.anthropic.com/v1/oauth/token",
     "api_base_url": "https://api.anthropic.com",
     "client_id": "9d1c250a-e61b-44d9-88ed-5944d1962f5e",
     "scope": "org:create_api_key user:profile user:inference",

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/plugins/claude_code_oauth/SETUP.md

@@ -37,7 +37,7 @@ Anthropic exposes a shared **public client** (`claude-cli`) for command-line too
 ```python
 CLAUDE_CODE_OAUTH_CONFIG = {
     "auth_url": "https://claude.ai/oauth/authorize",
-    "token_url": "https://claude.ai/api/oauth/token",
+    "token_url": "https://console.anthropic.com/v1/oauth/token",
     "api_base_url": "https://api.anthropic.com",
     "client_id": "9d1c250a-e61b-44d9-88ed-5944d1962f5e",
     "scope": "org:create_api_key user:profile user:inference",

{code_puppy-0.0.371 → code_puppy-0.0.373}/code_puppy/plugins/claude_code_oauth/utils.py

@@ -21,10 +21,10 @@ from .config import (
     get_token_storage_path,
 )
 
-# Proactive refresh buffer: refresh tokens 1 hour before expiration
-# This ensures smooth operation by refreshing during model requests,
-# well before the token actually expires
-TOKEN_REFRESH_BUFFER_SECONDS = 3600  # 1 hour
+# Proactive refresh buffer default (seconds). Actual buffer is dynamic
+# based on expires_in to avoid overly aggressive refreshes.
+TOKEN_REFRESH_BUFFER_SECONDS = 300
+MIN_REFRESH_BUFFER_SECONDS = 30
 
 logger = logging.getLogger(__name__)
 
@@ -146,15 +146,40 @@ def _calculate_expires_at(expires_in: Optional[float]) -> Optional[float]:
         return None
 
 
-def is_token_expired(tokens: Dict[str, Any]) -> bool:
+def _calculate_refresh_buffer(expires_in: Optional[float]) -> float:
+    default_buffer = float(TOKEN_REFRESH_BUFFER_SECONDS)
+    if expires_in is None:
+        return default_buffer
+    try:
+        expires_value = float(expires_in)
+    except (TypeError, ValueError):
+        return default_buffer
+    return min(default_buffer, max(MIN_REFRESH_BUFFER_SECONDS, expires_value * 0.1))
+
+
+def _get_expires_at_value(tokens: Dict[str, Any]) -> Optional[float]:
     expires_at = tokens.get("expires_at")
     if expires_at is None:
-        return False
+        return None
     try:
-        expires_at_value = float(expires_at)
+        return float(expires_at)
     except (TypeError, ValueError):
+        return None
+
+
+def _is_token_actually_expired(tokens: Dict[str, Any]) -> bool:
+    expires_at_value = _get_expires_at_value(tokens)
+    if expires_at_value is None:
+        return False
+    return time.time() >= expires_at_value
+
+
+def is_token_expired(tokens: Dict[str, Any]) -> bool:
+    expires_at_value = _get_expires_at_value(tokens)
+    if expires_at_value is None:
         return False
-    return time.time() >= expires_at_value - TOKEN_REFRESH_BUFFER_SECONDS
+    buffer_seconds = _calculate_refresh_buffer(tokens.get("expires_in"))
+    return time.time() >= expires_at_value - buffer_seconds
 
 
 def update_claude_code_model_tokens(access_token: str) -> bool:
@@ -216,11 +241,12 @@ def refresh_access_token(force: bool = False) -> Optional[str]:
             new_tokens = response.json()
             tokens["access_token"] = new_tokens.get("access_token")
             tokens["refresh_token"] = new_tokens.get("refresh_token", refresh_token)
-            if "expires_in" in new_tokens:
-                tokens["expires_in"] = new_tokens["expires_in"]
-                tokens["expires_at"] = _calculate_expires_at(
-                    new_tokens.get("expires_in")
-                )
+            expires_in_value = new_tokens.get("expires_in")
+            if expires_in_value is None:
+                expires_in_value = tokens.get("expires_in")
+            if expires_in_value is not None:
+                tokens["expires_in"] = expires_in_value
+                tokens["expires_at"] = _calculate_expires_at(expires_in_value)
             if save_tokens(tokens):
                 update_claude_code_model_tokens(tokens["access_token"])
                 return tokens["access_token"]
@@ -249,6 +275,11 @@ def get_valid_access_token() -> Optional[str]:
         refreshed = refresh_access_token()
         if refreshed:
             return refreshed
+        if not _is_token_actually_expired(tokens):
+            logger.warning(
+                "Claude Code token refresh failed; using existing access token until expiry"
+            )
+            return access_token
         logger.warning("Claude Code token refresh failed")
         return None
 

{code_puppy-0.0.371 → code_puppy-0.0.373}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "code-puppy"
-version = "0.0.371"
+version = "0.0.373"
 description = "Code generation agent"
 readme = "README.md"
 requires-python = ">=3.11,<3.14"