sapper-iq 1.0.16 → 1.0.18

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sapper-iq",
-  "version": "1.0.16",
+  "version": "1.0.18",
   "description": "AI-powered development assistant that executes commands and builds projects",
   "main": "sapper.mjs",
   "bin": {

package/sapper.mjs

@@ -132,29 +132,38 @@ async function runSapper() {
   if (messages.length === 0) {
     messages = [{
       role: 'system',
-      content: `You are Sapper, a senior engineer working in: ${process.cwd()}
+      content: `You are Sapper, a senior engineer.
 
 CRITICAL: You are working in the CURRENT DIRECTORY. Always use relative paths!
 - Use . or ./ for current directory
 - NEVER use / (that's the root directory)
 - Use relative paths like ./file.js or subfolder/file.js
 
-STRATEGY:
-1. When asked to analyze, use [TOOL:LIST].[/TOOL] first (NOTE: dot, not slash!)
-2. Read ONLY 2-3 KEY files in the SAME turn (README, package.json, main entry)
-3. DON'T read ALL files at once - read strategically!
-4. After reading files, PROVIDE ANALYSIS - don't just list more!
-5. Use format: [TOOL:TYPE]path]content[/TOOL]
+STRATEGY FOR FILE READING:
+1. Start with [TOOL:LIST].[/TOOL] to see what exists
+2. READ FILES BASED ON TASK:
+   - Quick overview: Read 2-3 key files (README, package.json, main entry)
+   - Deep analysis: Read ALL relevant files (entire src/ folder, all components)
+   - User asks "read all": Read ALL files they mention
+3. Use format: [TOOL:TYPE]path]content[/TOOL]
+4. After reading, PROVIDE ANALYSIS - don't just list more!
+
+READING GUIDELINES:
+- If user says "analyze src folder" → Read ALL files in src/
+- If user says "read everything" → List directory, then read all files
+- If < 20 files total: Read them all
+- If > 20 files: Ask user which area to focus on OR read all if analyzing entire project
+- If > 50 files: Read key files first, then ask user for focus area
 
 TOOL FORMAT (CRITICAL - FOLLOW EXACTLY):
 ✅ CORRECT: [TOOL:LIST].[/TOOL]
 ✅ CORRECT: [TOOL:READ]./file.js[/TOOL]
+✅ CORRECT: [TOOL:LIST]./src[/TOOL] then read all files found
 ❌ WRONG: [TOOL:LIST].[/] - missing TOOL at end!
 ❌ WRONG: [TOOL:LIST]/[/TOOL] - wrong directory!
 
 WORKFLOW:
-1. LIST directory → 2. READ 2-3 key files → 3. ANALYZE and RESPOND
-Don't keep executing tools endlessly - provide insights after reading!`
+1. LIST directory → 2. READ files (as many as needed) → 3. ANALYZE and RESPOND`
     }];
   }
 
@@ -212,27 +221,29 @@ Don't keep executing tools endlessly - provide insights after reading!`
           }
           fs.writeFileSync(CONTEXT_FILE, JSON.stringify(messages));
           
-          // Limit tool executions - if too many, warn and exit
-          if (toolMatches.length > 10) {
-            console.log(chalk.yellow('\n⚠️  Too many tool calls in one response! AI should analyze, not just read endlessly.'));
-            active = false;
+          // Warn if reading too many files at once
+          if (toolMatches.length > 30) {
+            console.log(chalk.yellow('\n⚠️  Reading 30+ files! This might take time. Consider being more selective.'));
           }
         } else {
-          // No tools found - check if malformed command
+          // No tools - check for malformed commands
           if (msg.includes('[TOOL:') && msg.includes('[/]')) {
-            console.log(chalk.red('\n❌ Malformed tool command detected! Expected format: [TOOL:TYPE]path[/TOOL]'));
+            console.log(chalk.red('\n❌ Malformed tool: Use [TOOL:TYPE]path[/TOOL]'));
             messages.push({ 
               role: 'user', 
-              content: 'ERROR: Your tool command is malformed. Use [TOOL:TYPE]path]content[/TOOL] or [TOOL:TYPE]path[/TOOL]' 
+              content: 'ERROR: Tool format wrong! Use [TOOL:TYPE]path[/TOOL]' 
             });
           } else {
-            // Normal response without tools - save context and wait for next input
+            // Normal response - save and continue
             fs.writeFileSync(CONTEXT_FILE, JSON.stringify(messages));
             active = false;
           }
         }
       }
       ask();
+    }).catch((error) => {
+      console.error(chalk.red('\n❌ Error in conversation loop:'), error);
+      ask(); // Continue despite error
     });
   };
   ask();