polydev-ai 1.8.13 → 1.8.15

package/lib/cliManager.js

@@ -183,6 +183,7 @@ class CLIManager {
       }
 
       let authenticated = false;
+      let quotaExhausted = false;
       
       // For Claude Code, skip command-based auth check and use file-based detection directly
       // This avoids the recursion issue when running from within Claude Code
@@ -211,6 +212,12 @@ class CLIManager {
           
           authenticated = this.parseAuthenticationStatus(provider.id, authOutput);
           
+          // Check for quota exhaustion (authenticated but rate limited)
+          if (provider.id === 'gemini_cli') {
+            quotaExhausted = authOutput.includes('exhausted your daily quota') ||
+                            authOutput.includes('quota') && authOutput.includes('exhausted');
+          }
+          
         } catch (authError) {
           if (process.env.POLYDEV_CLI_DEBUG) {
             console.log(`[CLI Debug] Auth check failed for ${provider.id}:`, authError);
@@ -262,6 +269,7 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       return {
         available: true,
         authenticated,
+        quota_exhausted: quotaExhausted,
         version,
         path: cliPath,
         last_checked: new Date(),
@@ -437,6 +445,19 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
         };
       }
 
+      // Check if CLI has quota exhausted - skip and suggest API fallback
+      if (providerStatus.quota_exhausted) {
+        console.log(`[Polydev CLI] ${provider.name} has exhausted daily quota, skipping CLI (use API fallback)`);
+        return {
+          success: false,
+          error: `${provider.name} has exhausted daily quota. Use API fallback.`,
+          error_code: 'QUOTA_EXHAUSTED',
+          latency_ms: Date.now() - startTime,
+          provider: providerId,
+          timestamp: new Date()
+        };
+      }
+
       // Log model being used
       if (model) {
         console.log(`[Polydev CLI] Using model for ${providerId}: ${model}`);
@@ -611,6 +632,22 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
           }
 
           lastErrorMessage = result.error;
+          
+          // Check for quota exhaustion error during execution
+          const combinedOutput = ((result.stdout || '') + ' ' + (result.stderr || '') + ' ' + (result.error || '')).toLowerCase();
+          if (combinedOutput.includes('exhausted') && combinedOutput.includes('quota') ||
+              combinedOutput.includes('rate limit') ||
+              combinedOutput.includes('too many requests')) {
+            console.log(`[Polydev CLI] ${providerId} quota/rate limit error detected during execution`);
+            return {
+              success: false,
+              error: `${provider.name} quota/rate limit exceeded. Use API fallback.`,
+              error_code: 'QUOTA_EXHAUSTED',
+              latency_ms: Date.now() - startTime,
+              provider: providerId,
+              timestamp: new Date()
+            };
+          }
         } catch (error) {
           lastErrorMessage = error instanceof Error ? error.message : String(error);
           

package/mcp/stdio-wrapper.js

@@ -100,6 +100,75 @@ if (writableTmp) {
   console.error(`[Stdio Wrapper] Using TMP directory: ${writableTmp}`);
 }
 
+/**
+ * Clean CLI response by removing metadata and debug info
+ * Strips: provider info, approval, sandbox, reasoning, session id, MCP errors, etc.
+ */
+function cleanCliResponse(content) {
+  if (!content || typeof content !== 'string') {
+    return content || '';
+  }
+  
+  const lines = content.split('\n');
+  const cleanedLines = [];
+  let skipUntilContent = true;
+  let inThinkingSection = false;
+  
+  for (const line of lines) {
+    const trimmed = line.trim();
+    
+    // Skip metadata lines at the start
+    if (skipUntilContent) {
+      // Skip empty lines
+      if (!trimmed) continue;
+      
+      // Skip known metadata patterns
+      if (trimmed.startsWith('provider:')) continue;
+      if (trimmed.startsWith('approval:')) continue;
+      if (trimmed.startsWith('sandbox:')) continue;
+      if (trimmed.startsWith('reasoning effort:')) continue;
+      if (trimmed.startsWith('reasoning summaries:')) continue;
+      if (trimmed.startsWith('session id:')) continue;
+      if (trimmed.match(/^-{4,}$/)) continue; // Skip separator lines like --------
+      if (trimmed === 'user') continue; // Skip role markers
+      if (trimmed === 'assistant') continue;
+      
+      // Skip MCP client errors
+      if (trimmed.startsWith('ERROR: MCP client for')) continue;
+      if (trimmed.includes('failed to start')) continue;
+      
+      // Found actual content - stop skipping
+      skipUntilContent = false;
+    }
+    
+    // Handle thinking sections (optional: can show or hide)
+    if (trimmed === 'thinking') {
+      inThinkingSection = true;
+      continue; // Skip the "thinking" marker
+    }
+    
+    // End of thinking section on empty line after content
+    if (inThinkingSection && !trimmed) {
+      inThinkingSection = false;
+      continue;
+    }
+    
+    // Skip more MCP errors anywhere in output
+    if (trimmed.startsWith('ERROR: MCP client for')) continue;
+    if (trimmed.includes('failed to start') && trimmed.includes('MCP')) continue;
+    
+    // Keep this line
+    cleanedLines.push(line);
+  }
+  
+  // Trim trailing empty lines and return
+  while (cleanedLines.length > 0 && !cleanedLines[cleanedLines.length - 1].trim()) {
+    cleanedLines.pop();
+  }
+  
+  return cleanedLines.join('\n');
+}
+
 class StdioMCPWrapper {
   constructor() {
     this.userToken = process.env.POLYDEV_USER_TOKEN;
@@ -559,6 +628,11 @@ class StdioMCPWrapper {
       for (const providerId of priorityOrder) {
         const status = results[providerId];
         if (status && status.available && status.authenticated) {
+          // Skip providers with quota exhausted - they'll use API fallback
+          if (status.quota_exhausted) {
+            console.error(`[Stdio Wrapper] Skipping ${providerId} - quota exhausted, will use API fallback`);
+            continue;
+          }
           availableProviders.push(providerId);
         }
       }
@@ -832,7 +906,7 @@ class StdioMCPWrapper {
           ? cliResult.model_used 
           : cliResult.provider_id;
         formatted += `🟢 **Local CLI Response** (${modelDisplay})\n\n`;
-        formatted += `${cliResult.content}\n\n`;
+        formatted += `${cleanCliResponse(cliResult.content)}\n\n`;
         formatted += `*Latency: ${cliResult.latency_ms || 0}ms | Tokens: ${cliResult.tokens_used || 0}*\n\n`;
         formatted += `---\n\n`;
       }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polydev-ai",
-  "version": "1.8.13",
+  "version": "1.8.15",
   "description": "Agentic workflow assistant with CLI integration - get diverse perspectives from multiple LLMs when stuck or need enhanced reasoning",
   "keywords": [
     "mcp",