Feature/implement conductor generation (#76)

* Initial toggleable implementation

* Remove sequential generator, conductor toggle

* Move simple chat, check memory in one place

* Minor cleanups

* Don't add user message twice

* Be a lot smarter about tool detection

* Remove bogus model settings

* Try to fix build errors

* Try again

* PR comment fixes and cleanup

* UTs

* Add thinking state and improve prompts

* Something more

* Testing with some logs

* Removed initial phase from conductor WIP

* WIP testing 3 phase approach

* Testing with Phase 1+2 combined

* Test fixes and linting

---------

Co-authored-by: Rishabh Chawla <rishabhchawla1995@gmail.com>
Co-authored-by: Pulkit Juneja <pulkitjuneja2@gmail.com>

Author: 3 people

package-lock.json

@@ -2,16 +2,17 @@ import * as readline from 'readline';
 import { queryEngineInstance } from './query_engine';
 import { RequestContext } from './logger';
 import  { ChatHistory } from './chat_history';
+import log from 'electron-log/main';
 
 async function main() {
     const rl = readline.createInterface({
       input: process.stdin,
       output: process.stdout,
     });
 
-    console.log("Chat with AI (type 'exit' to quit)");
-    console.log("Available modes: chat, contextaware");
-    console.log("Usage: /mode <mode> to switch modes (e.g. /mode contextaware)");
+    log.info("Chat with AI (type 'exit' to quit)");
+    log.info("Available modes: chat, contextaware");
+    log.info("Usage: /mode <mode> to switch modes (e.g. /mode contextaware)");
 
     const askQuestion = (): Promise<string> => {
       return new Promise((resolve) => {
@@ -21,15 +22,15 @@ async function main() {
       });
     };
 
-    let currentMode: 'CHAT' | 'CONTEXT_AWARE' = 'CHAT';
+    let currentMode: 'CHAT' | 'CONTEXT_AWARE' = 'CONTEXT_AWARE';
     const chatHistory = new ChatHistory();
 
     try {
       for (;;) {
         const userInput = await askQuestion();
 
         if (userInput.toLowerCase() === "exit") {
-          console.log("Goodbye!");
+          log.info("Goodbye!");
           rl.close();
           break;
         }
@@ -38,10 +39,10 @@ async function main() {
           const mode = userInput.slice(6).toUpperCase();
           if (['CHAT', 'CONTEXT_AWARE'].includes(mode)) {
             currentMode = mode as 'CHAT' | 'CONTEXT_AWARE';
-            console.log(`Switched to ${currentMode} mode`);
+            log.info(`Switched to ${currentMode} mode`);
             continue;
           } else {
-            console.log("Invalid mode. Available modes: chat, contextaware");
+            log.info("Invalid mode. Available modes: chat, contextaware");
             continue;
           }
         }
@@ -55,23 +56,22 @@ async function main() {
 
         chatHistory.addUserMessage(userInput);
 
-        console.log("AI: ");
+        log.info("AI: ");
         let response = "";
         const stream = await queryEngineInstance.query(requestContext, currentMode);
 
         for await (const chunk of stream) {
           process.stdout.write(chunk);
           response += chunk;
         }
-        console.log("\n");
+        log.info("\n");
 
         chatHistory.addAssistantMessage(response);
       }
     } catch (error) {
-      console.error("An error occurred:", error);
+      log.error("An error occurred:", error);
       rl.close();
     }
   }
 
-  main().catch(console.error);
-  
+  main().catch(log.error);

release/app/package-lock.json

@@ -56,15 +56,44 @@ class ChatHistory {
 
   /**
    * Convert chat history to format used by Ollama LLM
-   * @returns Array of Message objects in Ollama format
+   * Removes execution event metadata from AI context
+   * @returns Array of Message objects in Ollama format with clean content
    */
   toOllamaMessages(): Message[] {
     return this.messages.map(message => ({
       role: message.role,
-      content: message.content
+      content: this.cleanContentForAI(message.content)
     }));
   }
 
+  /**
+   * Clean content for AI context - removes ALL execution metadata
+   * This prevents AI from learning how to fake execution patterns
+   */
+  private cleanContentForAI(content: string): string {
+    let cleaned = content;
+    
+    // AGGRESSIVE CLEANING: Remove ALL XML-like tags that start with these patterns
+    cleaned = cleaned.replace(/<execution_event\b[^>]*>.*?<\/execution_event>/gs, '');
+    cleaned = cleaned.replace(/<tool_call_position\b[^>]*>/g, '');
+    cleaned = cleaned.replace(/<tool_calls_update\b[^>]*>.*?<\/tool_calls_update>/gs, '');
+    cleaned = cleaned.replace(/<tool_calls_complete\b[^>]*>.*?<\/tool_calls_complete>/gs, '');
+    cleaned = cleaned.replace(/<tool_calls\b[^>]*>.*?<\/tool_calls>/gs, '');
+    
+    // Remove standalone closing tags that might be left behind
+    cleaned = cleaned.replace(/<\/(?:execution_event|tool_calls_update|tool_calls_complete|tool_calls|tool_call_position)>/g, '');
+    
+    // Remove any remaining XML-like execution metadata
+    cleaned = cleaned.replace(/<[^>]*(?:execution|tool_call|metadata|event)[^>]*>.*?<\/[^>]+>/gs, '');
+    
+    // Clean up any excessive whitespace left behind
+    cleaned = cleaned.replace(/\n\s*\n\s*\n/g, '\n\n');
+    cleaned = cleaned.replace(/^\s+|\s+$/gm, ''); // Trim each line
+    cleaned = cleaned.trim();
+    
+    return cleaned;
+  }
+
   /**
    * Clear the chat history
    */