polydev-ai 1.8.9 → 1.8.11

package/lib/cliManager.js

@@ -443,38 +443,76 @@ 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 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: ${model || 'none'})`);
+        
+        // Try with specified model first, fallback to CLI default if model fails
+        let modelToUse = model;
+        let attempts = 0;
+        const maxAttempts = model ? 2 : 1; // Only retry if model was specified
+        
+        while (attempts < maxAttempts) {
+          attempts++;
+          try {
+            const result = await this.executeCodexExec(provider.command, execArgs, prompt, timeoutMs, modelToUse);
+            // executeCodexExec now returns { content, detectedModel, rawStdout, rawStderr }
+            const content = result.content;
+            const detectedModel = result.detectedModel;
+            
+            // Check if the response contains a model error
+            const modelError = content && (
+              content.includes('model is not supported') ||
+              content.includes('model not found') ||
+              content.includes('invalid model') ||
+              content.includes("doesn't exist")
+            );
+            
+            if (modelError && modelToUse && attempts < maxAttempts) {
+              console.log(`[Polydev CLI] Model '${modelToUse}' failed for Codex CLI, retrying with CLI default...`);
+              modelToUse = null; // Retry without model flag
+              continue;
+            }
+            
+            // Use detected model if available, otherwise fall back to what was requested or 'cli_default'
+            const actualModel = detectedModel || modelToUse || 'cli_default';
+            
+            if (detectedModel && detectedModel !== model) {
+              console.log(`[Polydev CLI] Codex CLI detected model: ${detectedModel} (requested: ${model || 'none'})`);
+            }
+            
+            return {
+              success: true,
+              content,
+              tokens_used: this.estimateTokens(prompt + content),
+              latency_ms: Date.now() - startTime,
+              provider: providerId,
+              mode: 'args',
+              model_used: actualModel,
+              timestamp: new Date()
+            };
+          } catch (error) {
+            const errorMsg = error instanceof Error ? error.message : String(error);
+            
+            // Check if error is model-related and we can retry
+            const isModelError = errorMsg.includes('model') && (
+              errorMsg.includes('not supported') ||
+              errorMsg.includes('not found') ||
+              errorMsg.includes('invalid')
+            );
+            
+            if (isModelError && modelToUse && attempts < maxAttempts) {
+              console.log(`[Polydev CLI] Model '${modelToUse}' error for Codex CLI, retrying with CLI default...`);
+              modelToUse = null; // Retry without model flag
+              continue;
+            }
+            
+            return {
+              success: false,
+              error: `CLI execution failed: ${errorMsg}`,
+              latency_ms: Date.now() - startTime,
+              provider: providerId,
+              mode,
+              timestamp: new Date()
+            };
           }
-          
-          return {
-            success: true,
-            content,
-            tokens_used: this.estimateTokens(prompt + content),
-            latency_ms: Date.now() - startTime,
-            provider: providerId,
-            mode: 'args',
-            model_used: actualModel,
-            timestamp: new Date()
-          };
-        } catch (error) {
-          return {
-            success: false,
-            error: `CLI execution failed: ${error instanceof Error ? error.message : String(error)}`,
-            latency_ms: Date.now() - startTime,
-            provider: providerId,
-            mode,
-            timestamp: new Date()
-          };
         }
       }
 
@@ -644,9 +682,50 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
 
       let stdout = '';
       let stderr = '';
+      let resolved = false;
+      let debounceTimer = null;
+
+      // Helper to check if output looks complete (for JSON output from claude code)
+      const looksComplete = () => {
+        const trimmed = stdout.trim();
+        // For JSON output, check if it's valid JSON
+        if (trimmed.startsWith('{') || trimmed.startsWith('[')) {
+          try {
+            JSON.parse(trimmed);
+            return true; // Valid JSON means complete
+          } catch {
+            return false;
+          }
+        }
+        // For text output, check if we have substantial content
+        return trimmed.length > 50 && trimmed.includes('\n');
+      };
+
+      const doResolve = () => {
+        if (resolved) return;
+        resolved = true;
+        if (debounceTimer) clearTimeout(debounceTimer);
+        if (timeoutId) clearTimeout(timeoutId);
+        if (!child.killed) {
+          try { child.kill('SIGTERM'); } catch(_) {}
+        }
+        resolve({ stdout, stderr });
+      };
+
+      // Schedule early return check after debounce period
+      const scheduleEarlyReturn = () => {
+        if (debounceTimer) clearTimeout(debounceTimer);
+        debounceTimer = setTimeout(() => {
+          if (!resolved && looksComplete()) {
+            console.log('[CLI Debug] Early return - output looks complete');
+            doResolve();
+          }
+        }, 2000); // Wait 2 seconds after last data for Claude Code (faster JSON parsing)
+      };
 
       child.stdout?.on('data', (data) => {
         stdout += data.toString();
+        scheduleEarlyReturn();
       });
 
       child.stderr?.on('data', (data) => {
@@ -659,6 +738,11 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       }
 
       child.on('close', (code) => {
+        if (resolved) return;
+        resolved = true;
+        if (debounceTimer) clearTimeout(debounceTimer);
+        if (timeoutId) clearTimeout(timeoutId);
+        
         if (process.env.POLYDEV_CLI_DEBUG) {
           console.log(`[CLI Debug] Command finished with code ${code}`);
         }
@@ -679,6 +763,10 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       });
 
       child.on('error', (error) => {
+        if (resolved) return;
+        resolved = true;
+        if (debounceTimer) clearTimeout(debounceTimer);
+        if (timeoutId) clearTimeout(timeoutId);
         if (process.env.POLYDEV_CLI_DEBUG) {
           console.log(`[CLI Debug] Command error:`, error);
         }
@@ -686,15 +774,11 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       });
 
       let timeoutId;
-      const cleanup = () => {
-        if (timeoutId) {
-          clearTimeout(timeoutId);
-          timeoutId = null;
-        }
-      };
 
       timeoutId = setTimeout(() => {
-        cleanup();
+        if (resolved) return;
+        resolved = true;
+        if (debounceTimer) clearTimeout(debounceTimer);
         if (!child.killed) {
           child.kill('SIGTERM');
           // Force kill after 2 seconds if still running
@@ -707,12 +791,8 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
         reject(new Error(`Command timeout after ${timeoutMs}ms`));
       }, timeoutMs);
 
-      child.on('close', () => {
-        cleanup();
-      });
-
       child.on('exit', () => {
-        cleanup();
+        if (debounceTimer) clearTimeout(debounceTimer);
       });
     });
   }
@@ -888,10 +968,12 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       let stdout = '';
       let stderr = '';
       let resolved = false;
+      let debounceTimer = null; // Smart debounce timer for early return
 
       const stop = (handler) => {
         if (!resolved) {
           resolved = true;
+          if (debounceTimer) clearTimeout(debounceTimer);
           try { child.kill('SIGTERM'); } catch (_) {}
           handler();
         }
@@ -981,18 +1063,31 @@ This is a known issue with @google/gemini-cli@0.3.4 and older Node.js versions.`
       // Check if we have a complete response - look for actual content
       const flushIfComplete = () => {
         const parsed = parseCodexOutput(stdout);
-        // Only resolve early if we have meaningful content (at least 2 chars) and output seems done
-        // This prevents cutting off multi-line responses
-        if (parsed && parsed.length >= 2 && stdout.includes('\n') && !stdout.endsWith('...')) {
+        // Only resolve early if we have meaningful content (at least 20 chars) and output looks complete
+        // Look for signs that Codex has finished outputting (tokens used, empty lines at end, etc.)
+        if (parsed && parsed.length >= 20) {
+          const detectedModel = this.detectModelFromOutput('codex_cli', stdout, stderr);
           clearTimeout(timeoutHandle);
-          stop(() => resolve(parsed));
+          if (debounceTimer) clearTimeout(debounceTimer);
+          stop(() => resolve({ content: parsed, detectedModel, rawStdout: stdout.trim(), rawStderr: stderr.trim() }));
         }
       };
 
+      // Smart debounce: wait 3 seconds after last data received before checking for early return
+      const scheduleEarlyReturn = () => {
+        if (debounceTimer) clearTimeout(debounceTimer);
+        debounceTimer = setTimeout(() => {
+          if (!resolved) {
+            console.log('[CLI Debug] Checking for early return after debounce...');
+            flushIfComplete();
+          }
+        }, 3000); // Wait 3 seconds after last data received
+      };
+
       child.stdout?.on('data', (data) => {
         stdout += data.toString();
-        // Don't flush too eagerly - wait a bit for more data
-        // flushIfComplete(); // Disabled: let the process complete naturally
+        // Schedule early return check after debounce period
+        scheduleEarlyReturn();
       });
 
       child.stderr?.on('data', (data) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polydev-ai",
-  "version": "1.8.9",
+  "version": "1.8.11",
   "description": "Agentic workflow assistant with CLI integration - get diverse perspectives from multiple LLMs when stuck or need enhanced reasoning",
   "keywords": [
     "mcp",