polydev-ai 1.8.3 → 1.8.5

package/lib/cliManager.js

@@ -46,7 +46,7 @@ class CLIManager {
           chat: [],
           version: ['--version'],
           auth_status: ['--print', 'test auth'], // Use --print to test auth
-          test_prompt: ['--print']
+          test_prompt: ['--print', '--output-format', 'json'] // Use JSON output to get model info
         },
         install_instructions: 'Install via: npm install -g @anthropic-ai/claude-code',
         auth_instructions: 'Authenticate with Claude Code'
@@ -444,7 +444,18 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       if (providerId === 'codex_cli') {
         const execArgs = promptVariants.find(args => args.includes('exec')) || promptVariants[0];
         try {
-          const content = await this.executeCodexExec(provider.command, execArgs, prompt, timeoutMs, model);
+          const result = await this.executeCodexExec(provider.command, execArgs, prompt, timeoutMs, model);
+          // executeCodexExec now returns { content, detectedModel, rawStdout, rawStderr }
+          const content = result.content;
+          const detectedModel = result.detectedModel;
+          
+          // Use detected model if available, otherwise fall back to what was requested or 'cli_default'
+          const actualModel = detectedModel || model || 'cli_default';
+          
+          if (detectedModel && detectedModel !== model) {
+            console.log(`[Polydev CLI] Codex CLI detected model: ${detectedModel} (requested: ${model || 'none'})`);
+          }
+          
           return {
             success: true,
             content,
@@ -452,7 +463,7 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
             latency_ms: Date.now() - startTime,
             provider: providerId,
             mode: 'args',
-            model_used: model || 'cli_default',
+            model_used: actualModel,
             timestamp: new Date()
           };
         } catch (error) {
@@ -499,7 +510,37 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
           );
 
           if (!result.error) {
+            // Special handling for Claude Code JSON output
+            if (providerId === 'claude_code') {
+              const jsonResult = this.parseClaudeCodeJsonResponse(result.stdout || '');
+              if (jsonResult) {
+                console.log(`[Polydev CLI] Claude Code detected model: ${jsonResult.model_used} (from JSON output)`);
+                return {
+                  success: true,
+                  content: jsonResult.content,
+                  tokens_used: jsonResult.tokens_used,
+                  latency_ms: Date.now() - startTime,
+                  provider: providerId,
+                  mode: 'args',
+                  model_used: jsonResult.model_used,
+                  cost_usd: jsonResult.cost_usd,
+                  model_usage: jsonResult.model_usage,
+                  timestamp: new Date()
+                };
+              }
+            }
+            
+            // Fallback: Standard text output handling
             const content = this.cleanCliResponse(result.stdout || '');
+            
+            // Detect actual model from CLI output
+            const detectedModel = this.detectModelFromOutput(providerId, result.stdout || '', result.stderr || '');
+            const actualModel = detectedModel || model || 'cli_default';
+            
+            if (detectedModel && detectedModel !== model) {
+              console.log(`[Polydev CLI] ${providerId} detected model: ${detectedModel} (requested: ${model || 'none'})`);
+            }
+            
             return {
               success: true,
               content,
@@ -507,7 +548,7 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
               latency_ms: Date.now() - startTime,
               provider: providerId,
               mode: 'args',
-              model_used: model || 'cli_default',
+              model_used: actualModel,
               timestamp: new Date()
             };
           }
@@ -531,6 +572,15 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
               
               if (!fallbackResult.error) {
                 const content = this.cleanCliResponse(fallbackResult.stdout || '');
+                
+                // Detect actual model from fallback output
+                const detectedModel = this.detectModelFromOutput(providerId, fallbackResult.stdout || '', fallbackResult.stderr || '');
+                const actualModel = detectedModel || 'cli_default_fallback';
+                
+                if (detectedModel) {
+                  console.log(`[Polydev CLI] ${providerId} fallback detected model: ${detectedModel}`);
+                }
+                
                 return {
                   success: true,
                   content,
@@ -538,7 +588,7 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
                   latency_ms: Date.now() - startTime,
                   provider: providerId,
                   mode: 'args',
-                  model_used: 'cli_default_fallback',
+                  model_used: actualModel,
                   timestamp: new Date()
                 };
               }
@@ -667,6 +717,113 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
     });
   }
 
+  // Detect actual model from CLI output before parsing/cleaning
+  detectModelFromOutput(providerId, stdout, stderr) {
+    const combinedOutput = (stdout || '') + '\n' + (stderr || '');
+    
+    // Different patterns for different CLI tools
+    switch (providerId) {
+      case 'claude_code': {
+        // Claude Code may output model info in various formats
+        // Look for patterns like "model: claude-sonnet-4-20250514" or "Using model: ..."
+        const claudeModelMatch = combinedOutput.match(/(?:model|using model)[:\s]+([a-zA-Z0-9_.-]+)/i);
+        if (claudeModelMatch && claudeModelMatch[1]) {
+          const model = claudeModelMatch[1].trim();
+          // Filter out invalid values
+          if (model && model !== 'undefined' && model !== 'null' && model.length > 2) {
+            return model;
+          }
+        }
+        break;
+      }
+      
+      case 'codex_cli': {
+        // Codex CLI outputs configuration header like:
+        // model:         gpt-4.1
+        // or: model: o4-mini
+        const codexModelMatch = combinedOutput.match(/^\s*model:\s*([a-zA-Z0-9_.-]+)/mi);
+        if (codexModelMatch && codexModelMatch[1]) {
+          const model = codexModelMatch[1].trim();
+          if (model && model !== 'undefined' && model !== 'null' && model.length > 2) {
+            return model;
+          }
+        }
+        break;
+      }
+      
+      case 'gemini_cli': {
+        // Gemini CLI may output model info
+        // Look for patterns like "model: gemini-2.0-flash" or "Using gemini-..."
+        const geminiModelMatch = combinedOutput.match(/(?:model|using)[:\s]+(gemini[a-zA-Z0-9_.-]*)/i);
+        if (geminiModelMatch && geminiModelMatch[1]) {
+          const model = geminiModelMatch[1].trim();
+          if (model && model.length > 2) {
+            return model;
+          }
+        }
+        break;
+      }
+    }
+    
+    return null; // Could not detect model from output
+  }
+
+  // Parse Claude Code JSON response to extract model and content
+  parseClaudeCodeJsonResponse(stdout) {
+    if (!stdout || !stdout.trim()) return null;
+    
+    try {
+      // Try to parse as JSON
+      const json = JSON.parse(stdout.trim());
+      
+      // Check if it's a valid Claude Code response
+      if (json.type !== 'result' || !json.result) {
+        return null;
+      }
+      
+      // Extract content
+      const content = json.result;
+      
+      // Extract primary model from modelUsage
+      // Priority: Opus > Sonnet > Haiku (prefer larger models as they do the main reasoning)
+      let primaryModel = 'cli_default';
+      const modelUsage = json.modelUsage || {};
+      const modelNames = Object.keys(modelUsage);
+      
+      if (modelNames.length === 1) {
+        // Only one model used
+        primaryModel = modelNames[0];
+      } else if (modelNames.length > 1) {
+        // Multiple models - prefer Opus > Sonnet > Haiku
+        const opusModel = modelNames.find(m => m.includes('opus'));
+        const sonnetModel = modelNames.find(m => m.includes('sonnet'));
+        const haikuModel = modelNames.find(m => m.includes('haiku'));
+        
+        primaryModel = opusModel || sonnetModel || haikuModel || modelNames[0];
+      }
+      
+      // Calculate total tokens
+      let totalTokens = 0;
+      for (const usage of Object.values(modelUsage)) {
+        totalTokens += (usage.inputTokens || 0) + (usage.outputTokens || 0) + 
+                       (usage.cacheReadInputTokens || 0) + (usage.cacheCreationInputTokens || 0);
+      }
+      
+      return {
+        content,
+        model_used: primaryModel,
+        tokens_used: totalTokens || json.usage?.input_tokens + json.usage?.output_tokens || 0,
+        cost_usd: json.total_cost_usd || 0,
+        model_usage: modelUsage,
+        session_id: json.session_id,
+        duration_ms: json.duration_ms
+      };
+    } catch (e) {
+      // Not valid JSON, return null to fall back to text parsing
+      return null;
+    }
+  }
+
   async executeCodexExec(executable, commandArgs, prompt, timeoutMs, model = null) {
     if (!executable) {
       throw new Error('Missing Codex executable');
@@ -849,18 +1006,21 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
         // Combine stdout and stderr for parsing - Codex may output to either
         const combinedOutput = trimmedStdout + '\n' + trimmedStderr;
 
+        // Detect the actual model used BEFORE parsing (model info is in the header)
+        const detectedModel = this.detectModelFromOutput('codex_cli', trimmedStdout, trimmedStderr);
+
         // Always try to parse stdout first, regardless of exit code
         // MCP handshake failures cause non-zero exit but don't prevent valid responses
         const parsedStdout = parseCodexOutput(trimmedStdout);
         if (parsedStdout) {
-          resolve(parsedStdout);
+          resolve({ content: parsedStdout, detectedModel, rawStdout: trimmedStdout, rawStderr: trimmedStderr });
           return;
         }
         
         // Try parsing combined output (some responses may appear in stderr)
         const parsedCombined = parseCodexOutput(combinedOutput);
         if (parsedCombined) {
-          resolve(parsedCombined);
+          resolve({ content: parsedCombined, detectedModel, rawStdout: trimmedStdout, rawStderr: trimmedStderr });
           return;
         }
 
@@ -868,7 +1028,7 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
         if (code === 0) {
           // Successful exit but no parseable output
           if (trimmedStdout) {
-            resolve(trimmedStdout); // Return raw output as fallback
+            resolve({ content: trimmedStdout, detectedModel, rawStdout: trimmedStdout, rawStderr: trimmedStderr }); // Return raw output as fallback
           } else {
             reject(new Error('Codex completed but produced no output'));
           }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polydev-ai",
-  "version": "1.8.3",
+  "version": "1.8.5",
   "description": "Agentic workflow assistant with CLI integration - get diverse perspectives from multiple LLMs when stuck or need enhanced reasoning",
   "keywords": [
     "mcp",
@@ -67,7 +67,7 @@
     "lucide-react": "^0.542.0",
     "marked": "^16.2.1",
     "next": "^15.5.7",
-    "polydev-ai": "latest",
+    "polydev-ai": "^1.8.4",
     "posthog-js": "^1.157.2",
     "prismjs": "^1.30.0",
     "react": "^18.3.1",