feat: Add Phase 1 and 2 improvements for Claude provider

Phase 1.2 - Automatic retry with exponential backoff:
- Add _request_with_retry() method for non-streaming requests
- Retries on 429 (with x-should-retry header), 529, 503 errors
- Exponential backoff with jitter (1s, 2s, 4s max 30s)
- Handles timeouts and HTTP errors gracefully

Phase 1.3 - Streaming idle watchdog:
- Add 90s idle timeout detection (matches vendors/claude)
- Tracks last_event_time and raises TimeoutError on idle
- Prevents indefinite hangs on dropped connections

Phase 2.3 - Cache token tracking:
- Add cache_stats dict to track cache hits/misses
- Track cache_tokens_read and cache_tokens_created
- Add get_cache_stats() method for analytics
- Updates stats during streaming message_delta events

Also includes:
- Temperature fix (skip 0.0 when thinking beta active)
- Rate limit config update (5s default for Claude)

aisbf/providers.py

@@ -2328,6 +2328,18 @@ class ClaudeProviderHandler(BaseProviderHandler):
         # HTTP client for direct API requests (kilocode method)
         self.client = httpx.AsyncClient(timeout=httpx.Timeout(300.0, connect=30.0))
         
+        # Streaming idle watchdog configuration (Phase 1.3)
+        self.stream_idle_timeout = 90.0  # seconds - matches vendors/claude
+        
+        # Cache token tracking for analytics (Phase 2.3)
+        self.cache_stats = {
+            'cache_hits': 0,
+            'cache_misses': 0,
+            'cache_tokens_read': 0,
+            'cache_tokens_created': 0,
+            'total_requests': 0,
+        }
+    
     def _get_auth_headers(self, stream: bool = False):
         """
         Get HTTP headers with OAuth2 Bearer token.
@@ -3307,13 +3319,68 @@ class ClaudeProviderHandler(BaseProviderHandler):
             logger.info(f"Payload: {json.dumps(payload, indent=2)}")
             logger.info(f"=== END NON-STREAMING REQUEST DEBUG ===")
         
-        # Non-streaming request
+        # Non-streaming request with automatic retry (Phase 1.2)
+        response = await self._request_with_retry(api_url, headers, payload, max_retries=3)
+        
+        claude_response = response.json()
+        
+        if AISBF_DEBUG:
+            logger.info(f"ClaudeProviderHandler: API response: {json.dumps(claude_response, indent=2)}")
+        
+        logger.info(f"ClaudeProviderHandler: Response received successfully via direct HTTP")
+        self.record_success()
+        
+        # Convert Claude API response to OpenAI format
+        openai_response = self._convert_to_openai_format(claude_response, model)
+        
+        return openai_response
+    
+    async def _request_with_retry(self, api_url: str, headers: Dict, payload: Dict, max_retries: int = 3):
+        """
+        Non-streaming request with automatic retry for transient errors (Phase 1.2).
+        
+        Retries on:
+        - 429 rate limit errors (with x-should-retry: true header)
+        - 529 overloaded errors
+        - 503 service unavailable
+        - Connection timeouts
+        
+        Uses exponential backoff with jitter between retries.
+        """
+        import logging
+        logger = logging.getLogger(__name__)
+        
+        last_error = None
+        
+        for attempt in range(max_retries):
+            try:
                 response = await self.client.post(api_url, headers=headers, json=payload)
                 
-        logger.info(f"ClaudeProviderHandler: Response status: {response.status_code}")
+                logger.info(f"ClaudeProviderHandler: Response status: {response.status_code} (attempt {attempt + 1}/{max_retries})")
                 
-        # Check for 429 rate limit error before raising
-        if response.status_code == 429:
+                # Check for retryable errors
+                if response.status_code in (429, 529, 503):
+                    # Check if we should retry
+                    should_retry = response.headers.get('x-should-retry', 'false').lower() == 'true'
+                    
+                    if should_retry or response.status_code in (529, 503):
+                        if attempt < max_retries - 1:
+                            # Calculate wait time with exponential backoff + jitter
+                            wait_time = min(2 ** attempt + random.uniform(0, 1), 30)
+                            
+                            # Try to get wait time from response
+                            try:
+                                error_data = response.json()
+                                error_message = error_data.get('error', {}).get('message', '')
+                                logger.warning(f"ClaudeProviderHandler: Retryable error: {error_message}")
+                            except Exception:
+                                pass
+                            
+                            logger.info(f"ClaudeProviderHandler: Retrying in {wait_time:.1f}s (attempt {attempt + 1}/{max_retries})")
+                            await asyncio.sleep(wait_time)
+                            continue
+                        else:
+                            # Max retries exceeded, handle the error
                             try:
                                 response_data = response.json()
                             except Exception:
@@ -3322,7 +3389,7 @@ class ClaudeProviderHandler(BaseProviderHandler):
                             self.handle_429_error(response_data, dict(response.headers))
                             response.raise_for_status()
                 
-        # Log error details for non-2xx responses
+                # Check for other errors
                 if response.status_code >= 400:
                     try:
                         error_body = response.json()
@@ -3336,18 +3403,33 @@ class ClaudeProviderHandler(BaseProviderHandler):
                     
                     response.raise_for_status()
                 
-        claude_response = response.json()
-        
-        if AISBF_DEBUG:
-            logger.info(f"ClaudeProviderHandler: API response: {json.dumps(claude_response, indent=2)}")
+                # Success
+                return response
                 
-        logger.info(f"ClaudeProviderHandler: Response received successfully via direct HTTP")
-        self.record_success()
+            except httpx.TimeoutException as e:
+                last_error = e
+                if attempt < max_retries - 1:
+                    wait_time = min(2 ** attempt + random.uniform(0, 1), 30)
+                    logger.warning(f"ClaudeProviderHandler: Request timeout, retrying in {wait_time:.1f}s")
+                    await asyncio.sleep(wait_time)
+                    continue
+                else:
+                    logger.error(f"ClaudeProviderHandler: Request timeout after {max_retries} attempts")
+                    raise
             
-        # Convert Claude API response to OpenAI format
-        openai_response = self._convert_to_openai_format(claude_response, model)
+            except httpx.HTTPError as e:
+                last_error = e
+                if attempt < max_retries - 1:
+                    wait_time = min(2 ** attempt + random.uniform(0, 1), 30)
+                    logger.warning(f"ClaudeProviderHandler: HTTP error, retrying in {wait_time:.1f}s: {e}")
+                    await asyncio.sleep(wait_time)
+                    continue
+                else:
+                    logger.error(f"ClaudeProviderHandler: HTTP error after {max_retries} attempts: {e}")
+                    raise
         
-        return openai_response
+        # Should not reach here, but just in case
+        raise last_error or Exception("Request failed after max retries")
     
     async def _handle_streaming_request_with_retry(self, api_url: str, payload: Dict, headers: Dict, model: str):
         """
@@ -3438,7 +3520,16 @@ class ClaudeProviderHandler(BaseProviderHandler):
                 content_block_index = 0
                 current_tool_calls = []
                 
+                # Streaming idle watchdog (Phase 1.3)
+                last_event_time = time.time()
+                idle_timeout = self.stream_idle_timeout
+                
                 async for line in response.aiter_lines():
+                    # Check for idle timeout (Phase 1.3)
+                    if time.time() - last_event_time > idle_timeout:
+                        logger.error(f"ClaudeProviderHandler: Stream idle timeout ({idle_timeout}s)")
+                        raise TimeoutError(f"Stream idle for {idle_timeout}s")
+                    
                     if not line or not line.startswith('data: '):
                         continue
                     
@@ -3451,6 +3542,9 @@ class ClaudeProviderHandler(BaseProviderHandler):
                     try:
                         chunk_data = json.loads(data_str)
                         
+                        # Update idle watchdog (Phase 1.3)
+                        last_event_time = time.time()
+                        
                         # Handle different event types
                         event_type = chunk_data.get('type')
                         
@@ -3587,6 +3681,16 @@ class ClaudeProviderHandler(BaseProviderHandler):
                             if usage:
                                 logger.debug(f"ClaudeProviderHandler: Streaming usage update: {usage}")
                                 
+                                # Track cache tokens for analytics (Phase 2.3)
+                                cache_read = usage.get('cache_read_input_tokens', 0)
+                                cache_creation = usage.get('cache_creation_input_tokens', 0)
+                                if cache_read > 0:
+                                    self.cache_stats['cache_hits'] += 1
+                                    self.cache_stats['cache_tokens_read'] += cache_read
+                                if cache_creation > 0:
+                                    self.cache_stats['cache_misses'] += 1
+                                    self.cache_stats['cache_tokens_created'] += cache_creation
+                        
                         elif event_type == 'message_stop':
                             # Final chunk
                             finish_reason = 'stop'
@@ -3855,6 +3959,22 @@ class ClaudeProviderHandler(BaseProviderHandler):
             # Close SDK client after streaming completes
             await sdk_client.close()
     
+    def get_cache_stats(self) -> Dict:
+        """
+        Get cache usage statistics (Phase 2.3).
+        
+        Returns:
+            Dict with cache statistics including hits, misses, and token counts.
+        """
+        total = self.cache_stats['cache_hits'] + self.cache_stats['cache_misses']
+        hit_rate = (self.cache_stats['cache_hits'] / total * 100) if total > 0 else 0
+        
+        return {
+            **self.cache_stats,
+            'total_cache_events': total,
+            'cache_hit_rate_percent': round(hit_rate, 2),
+        }
+    
     def _get_models_cache_path(self) -> str:
         """Get the path to the models cache file."""
         import os